We are IntechOpen, the world's leading publisher of Open Access books Built by scientists, for scientists

6,900+

Open access books available

186,000+

200M+ Downloads

International authors and editors

Our authors are among the

Top 1% most cited scientists

Contributors from top 500 universities

Selection of our books indexed in the Book Citation Index in Web of Science™ Core Collection (BKCI)

### Interested in publishing with us? Contact book.department@intechopen.com

Numbers displayed above are based on latest data collected. For more information visit www.intechopen.com

### IntechOpen Book Series Artificial Intelligence Volume 25

## Aims and Scope of the Series

Artificial Intelligence (AI) is a rapidly developing multidisciplinary research area that aims to solve increasingly complex problems. In today's highly integrated world, AI promises to become a robust and powerful means for obtaining solutions to previously unsolvable problems. This Series is intended for researchers and students alike interested in this fascinating field and its many applications.

## Meet the Series Editor

Andries Engelbrecht received the Masters and Ph.D. degrees in Computer Science from the University of Stellenbosch, South Africa, in 1994 and 1999 respectively. He is currently appointed as the Voigt Chair in Data Science in the Department of Industrial Engineering, with a joint appointment as Professor in the Computer Science Division, Stellenbosch University. Prior to his appointment at Stellenbosch University, he has been at the University of

Pretoria, Department of Computer Science (1998-2018), where he was appointed as South Africa Research Chair in Artifical Intelligence (2007-2018), the head of the Department of Computer Science (2008-2017), and Director of the Institute for Big Data and Data Science (2017-2018). In addition to a number of research articles, he has written two books, Computational Intelligence: An Introduction and Fundamentals of Computational Swarm Intelligence.

## Meet the Volume Editor

Yves Rybarczyk holds a Ph.D. in Robotics from the University of Evry, France. His teaching and research activities focus on artificial intelligence and human–machine interaction. He was an assistant professor at the Nova University of Lisbon between 2007 and 2015. Then, he moved to South America, where he was Associate Professor and Head of the Intelligent & Interactive Systems Lab, Universidad de Las Américas, Ecuador, until 2019. Currently, he is

a Full Professor of Data Analytics at Dalarna University, Sweden. He has participated in several projects on the modeling and development of complex and interactive systems. He is the author of more than 100 publications in scientific journals, book chapters, and conference papers. Besides his editorial activities in renowned journals, such as Frontiers in Big Data, Prof. Rybarczyk's work is recognized as pioneer research in the development of machine learning algorithms for predicting air quality.

### Contents



## Preface

Data mining has become a real challenge for modern societies. The boom of Big Data has allowed companies to obtain a more accurate and holistic evaluation of their business performance thanks to increasingly sophisticated analytic tools. We have moved on from simple spreadsheets to data lakes, which provide greater flexibility in identifying and summarizing a huge amount of data according to different business perspectives. Behind this optimization of data analysis, data mining has been enhanced with artificial intelligence and machine learning to extract patterns and create models to predict the future, thus becoming a valuable decision-making support.

This book presents a comprehensive overview of the new techniques and tools developed in data mining. Each section represents one of the three applications of data mining. Section 1, "Data Mining for Classification", includes three chapters. Chapter 1 proposes a model to find hidden connections between topics. This method is applied to medical concepts. Chapter 2 addresses the use of text mining to facilitate qualitative analyses in educational research. Chapter 3 is a study on the recognition of brain waves for classifying memory activities in rats.

Section 2, "Data Mining for Prediction", includes three chapters. Chapter 4 proposes a probabilistic method for revealing the most interesting if-then rules in the mined dataset. Chapter 5 discusses predicting COVID-19 lethality from several social and demographic factors. Chapter 6 presents a strategy to prevent the potential risk of adverse drug occurrences. This model is tested on heart attacks and thromboses.

Section 3, "Optimization Techniques Developed in Data Mining", includes three chapters. Chapter 7 describes a new method for the selection of the optimal power transformation parameters in regression analysis. Chapter 8 demonstrates the advantages of implementing a modified bagging to reduce the variance in classification error. Finally, Chapter 9 is an application study on the use of data mining for detecting bottlenecks in the production line and optimizing industrial processes.

This publication provides insight into recent methods and applications of data mining, analyzing its three main components: (1) classification, (2) prediction, and (3) optimization. These technical advancements are illustrated through pertinent applications in medicine, education, biology, health, and industry.

> **Yves Rybarczyk** School of Information and Engineering, Dalarna University, Falun, Sweden

Section 1

## Data Mining for Classification

#### **Chapter 1**

## Finding New Connections between Concepts from Medline Database Incorporating Domain Knowledge

*Yang Weikang, Chowdhury S.M. Mazharul Hoque and Jin Wei*

#### **Abstract**

In this digital world, data is everything and significantly impacts our everyday lives. Interestingly, in this small world, everything is part of an ecosystem, where everything is connected, directly or indirectly. The same thing happens to data as well. In most cases, it may seem like a particular topic does not have any connection with another one, but in reality, they are connected through a mutually related topic. Therefore, in this research, we will discuss an adaptive model modified from the ABC model by Don R. Swanson, a Literature-Based Discovery (LBD) Model, to find the hidden connections between Concepts of Interest. The model demonstrates that two topics, "A" and "C," are different and have no relationship. But they have a common topic, "B," that can be used to connect topics "A" and "C." This famous model will be used in this discussion to connect Medical Concepts.

**Keywords:** medical data analysis, NLP, Medline database, text extraction, web mining

#### **1. Introduction**

In this era of technology, the use of data is increasing rapidly, and with the growth of AI, data is growing even faster, as it requires a lot of data to come to a conclusion about a given problem. Throughout the world, the volume of scientific literature is expanding immensely. Because of that, it has become very challenging to keep track of the recent innovations, even for the specialized researchers with their slender domains. From all those literature-based contents, extracting useful information has become a challenge to everyone, and this can be solved using Literature-Based Discovery. LBD minimizes the difficulty among the researchers by reducing the complexity regarding finding useful information.

In this paper, we have worked on an extended version of the existing ABC model, and we applied the model to a medical research article database, the MEDLINE database. The reason behind this research was that every day, the medical sector is growing, and there are new problems that are coming up. There are a lot of existing problems that need to be solved, and researchers are working on them. As a result, every year, a huge number of articles are generated, and for anyone, keeping track of the data or the most recent research or research outcome would be a challenge. In the year 1999, a simple logic-based model was proposed by Swanson, known as

Complementary Structures in Disjoint Literature (CSD). In this model, he wanted to present that, in a research article, there can be many theories and findings that can be interconnected. For example, it is possible that in an article, it is mentioned that a certain medicine is used to treat the "Migraine Headache," and in another article, it says that that particular medicine commonly uses "Magnesium" as the main component [1]. Though "Migraine Headache" and "Magnesium" do not have a direct contact or relationship, and they exist in two different articles, it is still possible to make a relationship between those two.

This logic was explained by Swanson as follows – ""If concept A influences concept B, and concept B influences concept C, then concept A may influence concept C", and this model is also a Literature Based Discovery model called Swanson's ABC model [2]. Our research worked on an extended version of the ABC model. The model was applied to the Medline database, which is a database for worldwide medicalrelated research articles submitted to Medline and contains basic information, such as Title, Abstract, and Keywords. It is published by the National Library of Medicine (NLM). The proposed model uses the Pipe and Filter Software Architectural Design principles. The raw data was preprocessed to extract information that is relevant to the work before sending it to the final processing.

Moreover, the model was developed in such a way that it is not dependent only on two documents. Instead of that, it can go for cross-document discovery and produce a much better outcome. At the same time, the model was tested with a variety of cores to measure the impact on the result and processing time. It was found that the result on impact from the number of the cores was surprising. This model was tested and compared against other existing models and presented notable improvements. Data collected from the Medline database was processed using the MetaMap tool. It is a very popular tool that is used worldwide to process medical data, and it is also developed by NLM.

#### **2. Swanson's ABC model**

Don R. Swanson was the pioneer of Literature-Based Discovery, which is a form of knowledge extraction that can automatically generate hypotheses from a given dataset (basically a large amount of text data, such as abstracts of different scientific publications) to present new interdependence between existing concepts. The aim of LBD is to present indirect associations that are not totally direct. According to Don R. Swanson's hypothesis, two completely different studies that exhibit the A-B and B-C relationship may seem unchained, but an A-C relationship may emerge that was unexplored [3]. This process is also known as Swanson's Linking.

Don Swanson's original ABC model can be elaborated or explained through the "One Node Search" approach [4]. In this process, an initial topic or problem needs to be identified from an article that must have enough ground on the selected problem (Literature A). A second problem or topic needs to be selected from a second article with enough ground on the second topic and contains important information that would lead to the solution of problem A (Literature C). Here both articles have their own explanation and do not have any kind of interconnection. Now, from Literature A and C filtering important words and phrases needs to be carried out (B terms). B terms are the implicit information from both Literature A and C. Later, using those B terms (B1, B2, B3,………), searches need to be carried out to create B1, B2, B3….. literature from Literature A and C. By scoring each Bi literature from the title words and

*Finding New Connections between Concepts from Medline Database Incorporating Domain… DOI: http://dx.doi.org/10.5772/intechopen.113081*

phrases from A literature it needs to figure out how many B literatures they are in. To define A literature, each Ai term needs to be searched and analyzed. From the analysis, common Bi terms from literature A and literature C would hypothetically contain common information related to the solution of the given problem.

Through this process, it has become much easier and faster to analyze and inspect indirect connections to gain new knowledge that has not been explicitly stated. This technique showed significant progress, and similar models were used to make new breakthroughs.

#### **3. Literature review**

The use of semantic predictions for LBD was presented and advocated by Hristovski, who was among the first few people who understand its importance [5]. To recover existing knowledge and find new relationships, he used the pattern-based approach that leveraged both term co-occurrence from the BioMedLEE and semantics from the SemRep [6–9]. To find possible connections, he used a specified priori-based discovery pattern. However, this model's major drawback was it could not be easily extended to find complex patterns. Therefore, interesting and more sophisticated patterns may remain untouched.

Delory Cameron et al. worked on Swanson's hypothesis using a graph-based recovery and decomposition model using semantic predictions [5]. As biomedical literature-based semantic predictions allow the development of labeled directed graphs and from the literature, various concepts can be associated, a methodology for the semi-automatic discovery model was developed. Their model was a new approach to the field and was able to find associations at multiple levels.

Another study was done by Erwan Moreau et al. on the literature-based discovery of the ABC model to find the limitations of the ABC model [10]. Their work presents that the ABC model can find a relationship among the contexts, but how far it is effective or not effective is unknown to everyone, and the ABC model does not have the ability to judge it from a technical point of view. Therefore, they have used a data-driven task to work as a middle ground between the quantitative and qualitative learning-based discoveries. However, the model has an issue with the relatively small predefined targets.

Sam Henry et al. worked on the biomedical domain with an extensive literature review of literature-based discovery, where a unifying framework was selected, and different models and methods were differentiated and elaborated [11]. The purpose of the model was to give ideas to the readers about different LBD publications, models, methodologies, and challenges. Finally, with that knowledge, users will be able to develop an idea to create their own model.

Another work on the multilevel context-specific relationship discovery model of biological data was developed by Sejoon Lee et al. Their research goal was to find the relationship between drugs and diseases [12]. Their model proposed a three-step procedure to find the expected outcome, and the steps were multi-level entity identification, interaction extraction from the literature, and context vector-based similarity score calculation. The outcome from their model presented that context-based relation inference can perform better than the traditional ABC model.

Researcher Yong Hwan Kim focused on the limitations of the ABC model as it only gives priority to the B entity only and on the way to the B entity it may grab some relationships that are not entirely valid [13]. To counter the limitations, they presented a context-based approach to connect entities.

#### **4. Method**

As this model uses Pipe and Filter Architecture Design, multiple layers of filters work together to produce the final outcome from the raw data. Each filter works independently by taking input data from the previous filter in a processable format and producing the desired output format for the next filter. The whole analysis can be divided into three parts- preprocessing, preparation phase, and output phase. The whole process maintains a data flow that is given in **Figure 1**. The tool used to process data is the MetaMap, which is a popular tool that has great implementation in the text extraction and web mining field [14, 15].

#### **4.1 Data collection and preprocessing**

As mentioned earlier, all the data was collected from the Medline database through an API. This database is maintained by the NLM, and researchers across the world can get free access to research articles. At this moment, more than 23 million articles are stored in the Medline database. However, the basic type of data available in the database is the title of the article and abstract. Data were downloaded in XML format and from those XML files based on specific tags, information such as Publication ID, Publication Date, Article Title, and Article Abstract was extracted and stored. A total of 746 Medline documents were extracted from the website, which is - ftp://ftp.ncbi. nlm.nih.gov/pubmed/baseline. Each of the Medline documents has 30,000 Medline references, and each reference points towards a research document. Therefore, in total, around 22,380,000 references were analyzed by the system. All the documents that were collected were in English, and if any of the titles were in another language, they were translated into English first. **Figure 2** shows a sample of the collected raw data.

#### **4.2 Data processing through multiple filter**

In this proposed model, filtering was done through multiple modules, and each module works independently. Here, the first module is the MedMeta module and it initially monitors the preprocessing. Later, it forwards the preprocessed data to the MetaMap API, where the MetaMap breaks down all the sentences in the document into phrases. In this case, those sentences are the title and abstract of the research articles. It normalizes those phrases to find mapping concepts through its database search in order to assign a mapping score to the concepts. Based on those concepts

**Figure 1.** *Data flow diagram.*

*Finding New Connections between Concepts from Medline Database Incorporating Domain… DOI: http://dx.doi.org/10.5772/intechopen.113081*


#### **Figure 2.**

*Sample XML file containing publication date, article ID, title and abstract.*

that were generated by the MedMeta server, indexes of concept occurrence relationships are produced and passed to the next filter.

The processes from the beginning to the end are monitored by one of the modules that can be called the MedMeta class. It observes the creation of the objects and connects them to the other objects. This model uses a parser to control event occurrence depending on a certain threshold or situation and the main reason behind it is that the amount of data to be processed can reduce the effectiveness of the model. For example, the meline documents were around 97 gigabytes in size, and an individual document reached up to 200 megabytes. The use of a Java-based purser reduced that pressure and ensured maximum effectiveness.

Another module named MedlineCite was responsible for storing data that was produced after parsing from the previous module. It divided the output data from the previous module into four categories, such as ID, Title, Abstract, and Date.

The next part of the module is the most important one, as it is responsible for communication with the MetaMap server in order to build indexes. A code sample for this part of the module is given below in **Figure 3**.

Titles and Abstracts will be passed through the server as input data to be processed, and the server will use Word Sense Disambiguation to separate the phrases and ignore any stop phrases. After processing the data, the MetaMap server will return the output. The generated output will be forwarded to the next filter to continue the analysis.

#### *Research Advances in Data Mining Techniques and Applications*

#### **Figure 3.**

*Code snippet for connecting MetaMap server and index creation.*

After that, the next filter will take the output from the previous filter as input and will gather important information. A three-level data structure (semantic class, concept class, and occurrence class) will hold the collected information. The task of the semantic class will be holding one or many concept objects. The concept class will hold one or many occurrence objects. Finally, the occurrence class is responsible for representing the occurrence of a certain word in a Medline document. It will hold information such as the genuine word that appears in the text, the original word preferable name, word location in the citation, Title, Abstract, ID of the citation and publication date where the word appears, and the exact location of the word that appaired on the title abstract. A sample of the MetaMapped document is given below in **Figure 4**.

Now, this three-level data structure produces XLM type of output document to hold those generated data that is shown in the picture.

At this end of the processing MedMeta module uses a multithread based parallel processing in order to improve the performance. As mentioned earlier, each Medline document holds 30,000 citations and those citations are divided into three parts, each having 10,000 citations. Each part will be processed by a single thread, therefore, total three threads will be working together at the same time to access those data. Each thread will be able to communicate with the MetaMap server individually and will produce separate results. When all three of the threads are done with the analysis, all three threads will merge, and all the results will be forwarded to the main thread to start producing the output. Due to the use of multiple threads, it was possible to present the impact of using multithread against a single thread in the evaluation.

Here, another module was used named Semantic Type to Concept (S2C Module), which is responsible for generating a simplified version of the MetaMapped document or S2C document. Previously a three-level data structure was used to process data, and now, frequently, two-level S2C relationships will be used. There is a reason behind it. If every time a request is made and against that every time a three-level relationship is transformed into a two-level relationship, it will take a lot of time to process the data.

*Finding New Connections between Concepts from Medline Database Incorporating Domain… DOI: http://dx.doi.org/10.5772/intechopen.113081*

**Figure 4.** *Sample MetaMapped document.*

Therefore, by using a two-level S2C relationship, document processing time can be reduced. The S2C generator module starts its job by looking for the file path to pass them through the module as input, and a parser builds a two-level relationship. The parser basically uses an event-driven technique to go through all the MetaMapped documents to find concepts and include them in semantic-type classes. The semantic type class maintains a Hash set and only unique values will be added to the hash set. In this research, there were 133 semantic types, and individually they had a Hash set of concepts. Outcomes were stored in an XML file to be processed by the next module. A sample S2C document is presented in **Figure 5**.

A special note should be mentioned that, as only the title and abstract are playing the main role here, it is not necessary to keep anything other than those two in the main processing part. However, other relevant data can be used to know more about the source of the concept. Therefore, the Medline document will be more simplified and ready for concept generation. A sample of the simplified Medline document is shown in **Figure 6**.

The final module in this model is called the ClosedDiscovery module, and it uses the MetaMapped Document, Simplified Medline Document, S2C Document, and the user-provided keyword as input. Those inputs will be used to generate Semantic Type to Concept and Weight (S2CW), which is the final output of the model, and it will be displayed through a Graphical User Interface (GUI).

The graphical user interface has two input boxes to provide input to the system, basically, those inputs are the keywords- Topic A and Topic C, whom the user wants to connect. To start the process, the user needs to click the start button on the screen (given in **Figure 7**). In the next stage, users will be able to select the S2C document

**Figure 5.** *Sample S2C document.*

**Figure 6.**

*A simplified Medline document.*

#### **Figure 7.**

*GUI for the model (showing input boxes for topics A and B, and their concepts, including the outcome C).*

that they want to be analyzed. Other S2C documents will not be considered for processing. As soon as the processing begins, it will start looking for the sentences relevant to the given keywords and will be stored in two different XML files based on their keywords (Topic A and C) (**Figures 8**–**10**).

A concept chain will be created from those data by calculating the Term Frequency (TF) and the Inverse Document Frequency (IDF) in the sentences. The appearance of a concept in all sentences is considered as its Term Frequency. Multiple appearances of a concept in the same sentence will be considered as one appearance.

The Inverse Document Frequency measures the concept's importance. For concept C it is measured by –.

$$IDF\left(\mathbf{c}\right) = \text{loge}\left(\frac{\text{Total Number of Sentences}}{\text{Number of Sentences with concept C in it}}\right) \tag{1}$$

*Finding New Connections between Concepts from Medline Database Incorporating Domain… DOI: http://dx.doi.org/10.5772/intechopen.113081*


#### **Figure 8.**

*Page to selecting S2C documents for processing.*


#### **Figure 9.**

*GUI showing all the relevant sentences.*

In theory the minimal appearance of a concept in a sentence leads to high IDF value, making it important to the context. Later, the concept weight will be calculated by multiplying the TF and IDF, which is shown in Eq. (2).

$$\text{Weight}\,\text{C} = TFC^\ast \,\text{IDFC} \,\tag{2}$$

As the results from the MetaMap server may contain multiple concepts, therefore, by iterating them it is possible to build the Concept to Weight (C2W) and Concept

#### **Figure 10.**

*GUI presents sentences with intermediate linking concepts.*

to Sentence (C2Sents) relationship. Now the S2C relationship will be joined with the C2W relationship to find the S2CW relationship, and the weight of each concept is normalized by

$$\text{Normalized Weight} \left( \mathbf{c} \right) = \frac{\text{the weight of the concept C}}{\text{maximum weight in this semantic Type}} \qquad \text{(3)}$$

Now to get the intermediate level information S2CW for topic B (S2CWB), S2CW for topic A (S2CWA) and S2CW for topic C (S2CWC) will be merged together. Potential connecting terms between topic A and B will be represented by S2CWB by crisscrossing their concepts. That means, all the concepts in S2CWB will be present in both S2CWA and S2CWC. The weight of S2CWA and S2CWB will determine the weight of S2CWB. The resulting S2CWB will be the relationship between topic A and topic C. The data was presented in the GUI, the left side represents the concepts and references for Topic A, and the right side represents the concepts and references for Topic C. In the center, topic B, or the relationship between input topics are presented.

#### **4.3 System evaluation**

There were quite a few modifications during the development of the MedMeta module. One of the major modifications was changing the model from a single-thread to a multi-thread model. Therefore, each of those threads can independently communicate with the MetaMap server, and the performance significantly improved and processing time reduced, though CPU use increased. An experiment was done to know how far performance changes due to changes in the number of threads used

*Finding New Connections between Concepts from Medline Database Incorporating Domain… DOI: http://dx.doi.org/10.5772/intechopen.113081*


**Table 1.**

*Performance comparison on test data.*

to communicate with the MetaMap server. The performance comparison is shown in **Table 1** below. However, in this research, a computer was used that has a Core i7 4th generation CPU with eight cores, and 16 GB of RAM with Windows 10 64-bit OS. The document used for the test had 180 Medline citations. Testing was performed on 1 thread, 2 threads and 3 threads.

According to **Table 1**, performance improved greatly based on the increment of the number of threads. First of all, processing time was reduced to 145 s (1 thread) to 87 s when 2 threads were used and again dropped to 62 s when 3 threads were used. On the other hand, performance increased 1.67 times while it was 85% of the theoretical performance when 2 threads were used, and 2.34 times while it was 78% of the theoretical performance when 3 threads were used. However, due to hardware limitations and as running multiple MetaMap servers costs a lot, the number of threads could not be increased, but with a computer with better configuration, it can be improved more.

#### **4.4 Result analysis**

From the given data set, the model found some meaningful connecting concepts based on the input topics A and C. All the relevant information was stored in the S2CWB XML files. But to find the accuracy of the result, the model was compared with another model by Gopalakrishnan and Kishlay [16]. They have used their own model on Mesh terms related to discovery from given topics, and connecting mesh terms were discovered that explain the relationship of input topics and found to be meaningful. The same topics, such as Fish-oil and Raynaud's Disease, were given to this Medline Module with the same dataset to test the accuracy.

#### *4.4.1 Test set 1 - fish oil and Raynaud disease*

In the first test set by Gopalakrishnan and Kishlay was on 'Fish oil and Raynaud Disease'. In their experiment, they found keywords such as Arthritis Rheumatoid, Platelet Aggregation, Prostaglandin, Blood Viscosity, and Vascular Reactivity (**Table 2**).

The table presents that out of five connecting words, four were found by the model. On the other hand, three of the connecting words gained higher rank in their semantic type. This model used article title and abstract to find the linking contexts, but model by Gopalakrishnan and Kishlay used Mesh terms as potential connecting terms, that were assigned by biomedical experts. However, this model also found some linking terms that were undetected by their model, such as "Hemodynamic" and "Atherosclerosis".


#### **Table 2.**

*Analysis of fish oil and Raynaud disease.*

#### *4.4.2 Test set 2 - migraine and magnesium*

In the second study, two input topics were 'Migraine and Magnesium'. In the experiment of Gopalakrishnan and Kishlay, they found connecting words like, Serotonin, Norepinephrine, Propranolol, Calcium, Ergotamine, Adenine Nucleotides, Adenosine Triphosphate, and Epinephrine (**Table 3**).

For Migraine and Magnesium this model was able to find all eight connecting words from the Gopalakrishnan and Kishlay's model. Among those words, two of them got a very high semantic rank, and six of them were among the top ten semantic rank. Moreover, the model was able to find some connecting words, such as 'Insulin' that were not found previously.

#### *4.4.3 Test set 3 - schizophrenia and phospholipase A2*

In Gopalakrishnan and Kishlay's model 'Schizophrenia and Phospholipase A2' was used as the third test data and here the connecting words their model found are Trifluoperazine, Chlorpromazine, Prolactin, Choline, Norepinephrine, Arachidonic Acid, Receptors Dopamine and Phenothiazines (**Table 4**).

In this case study the result presented that all the connecting words from Gopalakrishnan and Kishlay's model were found again. Among eight of those words, three gained very high semantic score. Moreover, again, this model was able to find connecting words like, "PGE2", that were new to the model by Gopalakrishnan and Kishlay.


#### **Table 3.**

*Analysis on migraine and magnesium.*

*Finding New Connections between Concepts from Medline Database Incorporating Domain… DOI: http://dx.doi.org/10.5772/intechopen.113081*


#### **Table 4.**

*Analysis of schizophrenia and phospholipase A2.*

#### **5. Future work**

This model is currently discovering only one level of intermediate terms (chain length 1). In the future, our goal is to upgrade the model into multiple levels, so that when there is only a small amount of information available, it becomes easier to find possible connections. Moreover, MetaMapped files will be combined together for easier access to the files. Finally, other MetaMap options will be explored to find the one that can provide the best result.

#### **6. Conclusion**

This research discussed the Learning Based Discovery (LBD) models and, more specifically, one of the LBD models, which is the ABC model. An extended version of the ABC model was presented with a complete application. This era of technology is growing very fast, and with its growth, the use of data is also increasing. Therefore, sometimes, it is difficult to find relevant data about a particular topic. This complexity can be reduced by the use of LBD models. There is no LBD model that can be considered as the best model, because all the systems have their problems and benefits. A model may not provide the best result for a particular problem, but can perform amazingly for another. Moreover, interaction with the user interface makes any system easier. Because of that, the popularity of UI-based systems is increasing among researchers as well. However, this modified Swanson's ABC model can be improved more to solve problems faster and with a lot more data. This type of model will be able to make things easier for researchers by finding relevant data based on its user's input data from a larger data set.

#### **Acknowledgements**

This research is supported by the National Science Foundation award IIS-1739095.

#### **Author details**

Yang Weikang1 , Chowdhury S.M. Mazharul Hoque2 \* and Jin Wei2

1 North Dakota State University, Fargo, USA

2 University of North Texas, Denton, Texas, USA

\*Address all correspondence to: smmazharulhoquechowdhury@my.unt.edu

© 2023 The Author(s). Licensee IntechOpen. This chapter is distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/3.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

*Finding New Connections between Concepts from Medline Database Incorporating Domain… DOI: http://dx.doi.org/10.5772/intechopen.113081*

#### **References**

[1] Andreou AP, Edvinsson L. Mechanisms of migraine as a chronic evolutive condition. The Journal of Headache and Pain. 2019;**20**:117. DOI: 10.1186/s10194-019-1066-0

[2] Swason DR. Complementary structures in disjoint science literatures. In: Proceedings of the 14th Annual International ACM SIGIR Conference on Research and Development in Information Retrieval. Chicago, IL: ACM Press; 1991. pp. 280-289

[3] Swanson DR. Fish oil, Raynaud's syndrome, and undiscovered public knowledge. Perspectives in Biology and Medicine. 1986;**30**(1):7-18. DOI: 10.1353/ pbm.1986.0087

[4] Smalheiser NR. Rediscovering Don Swanson: The past, present and future of literature-based discovery. Journal of Data and Information Science (Warsaw, Poland). 2017;**2**(4):43. DOI: 10.1515/ jdis-2017-0019

[5] Cameron D, Bodenreider O, Hima Yalamanchili T, Danh SV, Thirunarayan K, Sheth AP, et al. A graph-based recovery and decomposition of Swanson's hypothesis using semantic predications. Journal of Biomedical Informatics. 2013;**46**(2):238-251, ISSN 1532-0464. DOI: 10.1016/j. jbi.2012.09.004

[6] Hristovski D, Friedman C, Rindflesch TC, Peterlin B. Exploiting semantic relations for literature-based discovery. In: AMIA Annual Symposium Proceedings. AMIA Symposium. National Library of Medicine; 2006. p. 349-53

[7] Hristovski D, Friedman C, Rindflesch TC, Peterlin B.

Literature-based knowledge discovery using natural language processing. In: Bruza P, Weeber M, editors. Literature-Based Discovery, Information Science and Knowledge Management. Vol. 15. Berlin, Heidelberg: Springer; 2008. pp. 133-152

[8] Rindflesch TC, Fiszman M. The interaction of domain knowledge and linguistic structure in natural language processing: Interpreting hypernymic propositions in biomedical text. Journal of Biomedical Informatics. 2003;**36**(6):462-477

[9] Lussier Y, Borlawsky T, Rappaport D, Liu Y, Friedman C. Phenogo: Assigning phenotypic context to gene ontology annotations with natural language processing. In: Pacific Symposium on Biocomputing. National Library of Medicine; 2006. p. 64-75

[10] Moreau E, Hardiman O, Heverin M, O'Sullivan D. Literature-based discovery beyond the ABC paradigm: A contrastive approach. BioRxiv. 2021;**09**(22):461375. DOI: 10.1101/2021.09.22.461375

[11] Henry S, McInnes BT. Literature based discovery: Models, methods, and trends. Journal of Biomedical Informatics. 2017;**74**:20-32, ISSN 1532-0464,. DOI: 10.1016/j.jbi.2017.08.011

[12] Lee S, Choi J, Park K, et al. Discovering context-specific relationships from biological literature by using multi-level context terms. BMC Medical Informatics and Decision Making. 2012;**12**(Suppl 1):S1. DOI: 10.1186/1472-6947-12-S1-S1

[13] Kim YH, Song M. A context-based ABC model for literature-based discovery. PLoS One. 2019;**14**(4):e0215313. DOI: 10.1371/ journal.pone.0215313

[14] Aronson AR. Effective mapping of biomedical text to the UMLS Metathesaurus: The MetaMap program. In: AMIA Annual Symposium Proceedings. AMIA Symposium. National Library of Medicine; 2001. p. 17-21. DOI: D010001275

[15] U.S. National Library of Medicine. MetaMap - A Tool for Recognizing UMLS Concepts in Text. 2017. Available from: https://metamap.nlm.nih.gov/

[16] Gopalakrishnan V, Jha K, Zhang A, Jin W. Generating hypothesis: Using global and local features in graph to discover new knowledge from medical literature. In: Proceedings of the 8th International Conference on Bioinformatics and Computational Biology, BICOB. USA; 2016. pp. 23-30

#### **Chapter 2**

## Artificial Intelligence in Educational Research

*Ulises Alejandro Duarte Velazquez*

#### **Abstract**

The proliferation of textual data in academic literature necessitates accelerating qualitative research methodologies. Text mining, underpinned by artificial intelligence and natural language processing, emerges as a transformative solution. This study analyzes how AI-integrated qualitative data analysis software such as ATLAS. ti and MAXQDA have streamlined processes like automatic coding and summarization since early 2023. These tools now facilitate rapid preliminary reviews through summarization features and obviate programming expertise through intuitive interfaces. Key advantages include drastic reductions in manual coding time through AI coding, enrichment of inductive coding systems via semantic analysis-based sub-code suggestions, and insights-driving code commenting summaries. Deep learning models unlocked by such tools will enable discernment of increasingly intricate patterns, improving educational interventions through real-time strategies informed by empirical findings. However, responsible use requires human oversight to refine coding and interpret nuanced results. While propelling qualitative research to unprecedented scales and depths, text mining also poses challenges around potential oversight neglect and lack of ethical guidelines. Optimizing these tools ensures accurate, responsible analyses that revolutionize understanding complex educational processes. AI ultimately enhances social science and education research outcomes through large-scale textual data analysis.

**Keywords:** text mining, qualitative data analysis, artificial intelligence, natural language processing, educational research

#### **1. Introduction**

The world's information is growing at an exponential rate, with estimates indicating that 80% of all information is in text form [1]. The academic landscape is undergoing a significant increase in the volume of scholarly publications [2, 3]. This proliferation is particularly impactful in the field of data science, as artificial intelligence can assist in generating novel research methodologies, especially when it comes to analyzing large volumes of textual information. This is further facilitated by the continual advancements in Natural Language Processing (NLP).

Text mining, which has its roots in artificial intelligence (AI), significantly reduces the time required for data analysis in qualitative research that relies on text analysis.

It employs statistical techniques and leverages algorithms to scrutinize written information [4, 5]. This technological advancement not only streamlines the research process but also enhances the depth and breadth of insights that can be gleaned from large textual datasets, given that it is impractical to manually analyze large volumes of text, text mining can uncover patterns that may not be immediately apparent to the researcher. The utilization of artificial intelligence has the potential to guide education towards the discovery of new curricula, teaching methods, and novel research avenues [6].

Text mining is a suite of processes that employs Natural Language Processing to facilitate human-machine communication, mediated by artificial intelligence; thus, a computer is essential for this process [7]. Text mining has found diverse applications throughout history, ranging from politics to medicine. For instance, it has been used to analyze behavioral patterns in patients through alcohol-related forums, employing Latent Dirichlet allocation (LDA) techniques [8]. Although its applications date back to the 1980s, in the educational field, text mining can be utilized to enhance teaching and learning processes, as well as to analyze issues related to educational underperformance or success [9]. By employing Quantitative Text Analysis (QTA), one can systematically analyze various collections of text in an automated manner, identifying underlying structures or patterns within the text. It is important to note that this approach is not intended to replace careful reading. Rather, its aim is to augment the understanding of the information, thereby enhancing the depth and rigor of textual analysis. This method serves as a complementary tool that can facilitate more nuanced interpretations and insights into the subject matter under investigation [10].

#### **2. Exploring educational data through text mining**

Furthermore, text mining can be employed to analyze student feedback, aiming for a more comprehensive understanding of their needs and experiences. It can also be used to review scholarly articles, with the objective of constructing clear maps of prevailing trends in the educational field. This capability not only enriches our understanding of the educational landscape but also provides actionable insights for targeted improvements [11]. Therefore, educational research should seize the new opportunities offered by AI-supported text mining to analyze large volumes of data, a task that was previously unattainable for individuals without expertise in programming. This opens a new frontier for educational innovation and data-driven policy formulation.

In the study titled "Artificial Intelligence and Reflections from Educational Landscape: A Review of AI Studies in Half a Century" [12] an increase in publications related to AI was observed. These publications show a clear concentration of such research endeavors in the United States and China, falling into three major clusters: AI, pedagogical issues, and technological matters. However, the study also notes that ethics, despite its importance, is a topic often overlooked in research involving artificial intelligence.

In the study titled "Text Mining Undergraduate Engineering Programs' Applications: The Role of Gender, Nationality, and Socio-economic Status" [13], the aim was to identify user behavior patterns related to Massive Open Online Courses (MOOCs). For this purpose, topic models with Latent Dirichlet allocation were employed. This approach assists in uncovering the hidden semantic structure within the analyzed texts:

*The process begins by drawing a K-dimensional Dirichlet vector θd that captures the expected proportion of words in document d that can be attributed to each topic. Then for each position (or, equivalently, for each word) in the document, indexed by n, it proceeds by sampling an indicator zd, n from a Multinomial K (θd, 1) whose positive component denotes which topic such position is associated with. The process ends by sampling the actual word indicator wd,n from a Multinomial V(Bzd,n, 1), where the matrix B = [β1|. . . |βK], encodes the distributions over terms in the vocabulary associated with the K topics [14].*

Topic modeling leverages machine learning techniques, wherein Natural Language Processing capitalizes on the presence of textual data to enable computers to understand words and learn data patterns [15]. This sophisticated approach not only facilitates the extraction of meaningful insights but also enhances the computational capabilities for analyzing large and complex textual datasets.

In the study titled "Data Mining Analysis of the 2022 Curriculum Framework and Study Plan for Basic Education in Mexico" [16] a comparison was made between the 2011, 2017, and 2022 Mexican Study Plans using text mining with packages such as Quanteda, Tidyverse, and word2vec. The study indicates that the 2022 plan promotes inclusion and well-being for populations facing inequality.

In the study titled "Learner-Centered Analysis in Educational Metaverse Environments: Exploring Value Exchange Systems through Natural Interaction and Text Mining" [17] text mining techniques were applied to a compilation of comments, discussions, and reflections. Utilizing term extraction, co-occurrence analysis, and network modeling, valuable information was gleaned regarding student learning outcomes. In conclusion, text mining proved to be a powerful tool for exploring self-directed learning within the ever-changing landscape of education in the digital age.

The study titled "Exploring the Advancements and Future Research Directions of Artificial Neural Networks: A Text Mining Approach" [18] examines the use of Artificial Neural Networks (ANN) through a text mining lens, utilizing a corpus of 10,661 articles and 35,973 keywords from scientific journals. The results indicate that research has evolved from optimization and machine learning towards deep learning and artificial intelligence. This progression has led to improved predictive models that are capable of understanding and modeling complex systems, including educational processes and teaching-learning mechanisms. This study serves as a comprehensive overview, shedding light on the transformative impact of ANN in various domains.

Educational research must embrace the challenges associated with utilizing artificial intelligence for data analysis, particularly for text-based data, to gain a deeper understanding of teaching and learning processes. The proliferation of technologies that leverage artificial intelligence consequently offers innovative opportunities for educational research. Previously, conducting such analyses required a researcher or a team member to possess expertise in programming and statistical methods for text mining. However, with the advent of deep learning models and neural networks, such specialized knowledge is no longer essential. This shift signifies the democratization of new research techniques, which are readily accessible in contemporary research software platforms.

#### **3. Possibilities of using research software with artificial intelligence**

Educational research must embrace the challenges associated with utilizing artificial intelligence for data analysis, particularly for text-based data, to gain a deeper understanding of teaching and learning processes. The proliferation of technologies that leverage artificial intelligence consequently offers innovative opportunities for educational research. Previously, conducting such analyses required a researcher or a team member to possess expertise in programming and statistical methods for text mining. However, with the advent of deep learning models and neural networks, such specialized knowledge is no longer essential. This shift signifies the democratization of new research techniques [19], which are readily accessible in contemporary research software platforms.

With natural language processing, research software designed to identify patterns in text is evolving to a new level of analysis. NLP allows for the examination of large volumes of textual data without the need for manual coding [20], a practice to which researchers using tools like MAXQDA, ATLAS.ti, or Quirkos had previously become accustomed.

Education is inherently more qualitative than quantitative in nature; thus, qualitative text analysis is most employed for this type of data [21–24]. Since the beginning of 2023, some software platforms like MAXQDA and ATLAS.ti have started to incorporate artificial intelligence to assist researchers in the processes of systematization and analysis. These platforms enable researchers to label text through coding, creating a comprehensive coding system that offers deep insights. This facilitates the construction of categories, which can be developed either from a theoretical perspective that relies more on human judgment or through an empirical approach [25]. The former represents a deductive analytical framework, while the latter adopts a more inductive stance, which is particularly well-suited for information management using AI.

#### **3.1 ATLAS.ti and its possibilities with artificial intelligence for data analysis**

ATLAS.ti is a Qualitative Data Analysis (QDA) software designed to manage and analyze large volumes of qualitative data, ranging from interview transcripts and researcher notes to documents, images, and videos. One of the earliest studies to utilize ATLAS.ti was "ATLAS/ti — A Prototype for the Support of Text Interpretation" [26], which described how researchers could create maps linking concepts in a network-like fashion. The software also allows for the writing of memos, thereby assisting researchers in maintaining organization through its graphical user interface. This comprehensive toolset not only streamlines the research process but also enhances the depth and quality of qualitative analyses.

Currently, ATLAS.ti leverages OpenAI technology to extract qualitative insights in a matter of minutes, transforming the landscape of qualitative data analysis. Whereas the traditional coding process could take weeks to complete, the software now promises to expedite research projects by up to tenfold [27] with the aid of OpenAI. It offers an array of advanced features such as automated coding, real-time code suggestions, automatic summarization, and data visualization tools, thereby revolutionizing the efficiency and effectiveness of qualitative research in both academic and professional settings.

As can be seen in **Figure 1** (screenshot captures), from the moment the user imports a document, ATLAS.ti enables the use of artificial intelligence to initiate document analysis. For demonstration purposes, Mexico's 2022 Curriculum Plan was used as a sample. The process is as straightforward as selecting the document to

#### *Artificial Intelligence in Educational Research DOI: http://dx.doi.org/10.5772/intechopen.113844*


#### **Figure 1.**

*Screenshot captures of ATLAS.ti.*

be analyzed and clicking the "Summarize" button, which then yields the following results. This streamlined approach significantly simplifies the qualitative analysis process, making it more accessible and efficient for researchers.

*ATLAS.ti results: "In conclusion, Mexico's Secretariat of Public Education (SEP) is implementing strategies to strengthen inclusive education, gender equity, and respect for human rights. The aim is to improve the management of educational resources and encourage community participation in enhancing education. The curricular proposal is grounded in respect for diversity and aims to cultivate citizens committed to social transformation. The importance of strengthening ties between schools and communities is recognized, and the professional autonomy of teachers is emphasized. The approach also seeks to foster holistic skills in students and underscores the importance of their involvement in shaping the educational environment. The New Mexican School promotes well-being and respectful treatment, acknowledges the importance of family participation, and aims to address digital inequalities while fostering comprehensive education in digital culture. During the pandemic, teachers adapted to working from home, and the importance of mathematics and other disciplines in everyday life was acknowledged. However, the curriculum has also been criticized for historically promoting a mestizo vision, thereby neglecting the country's ethnic and cultural diversity. Each educational reform is a political-pedagogical project that defines the curriculum and teaching methods based on a vision for the nation and the diverse groups within society."*

With ATLAS.ti, researchers have the capability to upload various research documents and obtain summaries. As demonstrated in the previous example, a summary was generated for a document exceeding 200 pages in length. This feature significantly streamlines the research process, allowing for quick and efficient overviews of extensive materials, thereby enhancing the researcher's ability to focus on key insights and findings.

**Figure 2.** *Screenshot capture of ATLAS.ti (IA codig).*

Similarly, the most significant advancement that accelerates the coding process for research is the implementation of automatic coding through artificial intelligence. In ATLAS.ti, this can be easily accomplished by following the "Search & Code → IA Code" pathway, as shown in **Figure 2**. Researchers simply need to select the document to be coded and then wait for the computer to complete the task. This innovation greatly enhances the efficiency and accuracy of the qualitative analysis process.

Thus, utilizing this process, 44 parent codes were generated from the document in just 30 minutes using a PC equipped with an Intel(R) Core(TM) i5-7400 CPU @ 3.00GHz and 8GB of RAM. These parent codes were accompanied by various child codes or sub-codes, amounting to a total of 1210 codings. Although the software is still in its beta stage, this number of codes can be overwhelming when the goal is to generate clear labels (codes) that identify thematic patterns. Therefore, human oversight is essential to refine the coding system by merging similar codes, deleting unnecessary ones, and thereby reducing the dataset to facilitate clearer categorization of information. Nevertheless, the use of artificial intelligence undoubtedly serves to dramatically reduce the amount of time required for this labor-intensive process, making it a valuable tool in modern qualitative research.

#### **3.2 MAXQDA and its possibilities with artificial intelligence for data analysis**

MAXQDA is another software platform that has embraced text mining using Natural Language Processing. Initially developed in Germany in 1998 to facilitate qualitative data analysis, the software was originally named WINMAX. It was designed following the methodologies of Max Weber and Alfred Schutz [28]. Over the years, MAXQDA has evolved to incorporate advanced text mining capabilities, thereby expanding its utility, and making it a versatile tool for researchers in various disciplines seeking to analyze complex textual data.

#### *Artificial Intelligence in Educational Research DOI: http://dx.doi.org/10.5772/intechopen.113844*

Currently, MAXQDA is designed for managing and analyzing qualitative data, but it also supports mixed-methods research. The software allows for the handling of interview transcriptions, field notes, documents, images, and videos. In its 2022 version, a new virtual research assistant feature, known as AI Assist, has been incorporated. This feature enables researchers to streamline their work by automatically analyzing and summarizing various elements of the research project. Notably, it integrates with OpenAI's AI technology, the creator of ChatGPT, and is currently in its beta stage.

The three primary tools available in MAXQDA are: (A) "Summarize Selected Text Passages," which condenses portions of text selected by the researcher and saves them directly as notes; (B) "Suggest Subcodes," where the software analyzes text portions based on codes developed by the researcher and offers suggestions for possible new sub-codes; and (C) "Summarize a Coded Segment to a Code Comment," which allows for the creation of summaries of coded segments that are saved as code comments. These features collectively enhance the software's utility, making it a comprehensive and versatile tool for academic research in various disciplines (**Figure 3**).

To utilize the AI Assist feature, researchers must first establish their own code system, underscoring the continued necessity for human oversight to achieve optimal results. This step allows for a more logical linkage of text segments with the corresponding codes. For illustrative purposes, the 2022 Mexican Study Plan was used as an example. The process involves the following steps: select code → right-click → AI Assist → Summarize Code Segments → choose language → accept policies and confirm. After a variable amount of time, depending on the length of the coded text segments, the software generates the results. This comprehensive approach enhances the reliability and depth of the analysis, making it a valuable tool for academic research. In this specific example, the code "learning" was used, yielding the following summary.

*Learning is fundamental in the New Mexican School, which aims to guarantee the right to education for children and youth. The community serves as the integrating core of teaching and learning processes, emphasizing the need to contextualize content to address the regional, local, contextual, and situational nature of the teaching and learning process. Additionally, there is criticism of the focus on cognitive capitalism and the knowledge society in education.*


#### **Figure 3.** *Screenshots capture of MAXQDA (IA Assit – Summarize coded segments).*

*The assessment of learning should be formative and centered on tracking the learning process of students. Accreditation should substantiate the granting of both grades and certificates based on the teachers' judgment of the entire teaching and learning process. The learning outcome profile is based on capacities, values, knowledge, and skills gradually acquired in formative fields. Articulating axes connect the content of different disciplines and teaching and learning actions with the students' reality.*

*The curricular proposal emphasizes the importance of the community as a social, cultural, political, productive, and symbolic space in which the school is situated. The socially constructed knowledge and wisdom of community life by the students should be considered to discuss and even problematize the content of the Study Plan and Programs. The community introduces into the school a living dialog between the present and the past, allowing students to see themselves as historical subjects.*

*Formative fields address the relationship of human beings with society and nature, the construction of participatory and responsible citizenship, and the building of identity and a sense of belonging. Study programs must be adapted to the concrete conditions of teaching and learning. Phases and grades of learning indicate the indispensable sequences between disciplines.*

*In summary, public education in Mexico focuses on student development through formative processes centered on the community and contextualized at the regional and local levels. The assessment is formative and based on tracking the students' learning process, and the exit profile is grounded in capacities, values, knowledge, and skills gradually acquired in formative fields. The curricular proposal underscores the importance of the community as a multifaceted space where the school is embedded, and formative fields explore human relationships with society and nature, participatory citizenship, and identity construction.*

To illustrate the "Suggest Subcodes" feature, the same document was used, specifically focusing on the "learning" code from the 2022 Mexican Study Plan. As depicted in **Figure 4**, the following procedure is followed: select code → right-click → AI Assist → Suggest Subcodes → choose language → accept policies and confirm. In this manner, MAXQDA, in collaboration with OpenAI, analyzes the information within


#### **Figure 4.**

*Screenshot capture of MAXQDA (IA Assit – Suggest subcodes).*

the code and its associated text segments to provide suggestions. The process culminates in generating the following results, thereby offering a nuanced and data-driven approach to qualitative research.

	- Designing programs of study to address regional, local, contextual, and situational aspects of teaching and learning.
	- Community as the integrative core of teaching and learning processes.
	- Contextualization of content for diverse learners and changing social, cultural, environmental, and health contexts.
	- Prioritizing maximum learning achievement for all students.
	- Developing integral human development of learners.
	- Promoting a culture of educational responsibility and social transformation.
	- Recognizing diversity as a condition and starting point for learning processes.
	- Valuing and incorporating cultural, social, digital, economic, and linguistic expressions of learners.
	- Fostering autonomy and professional decision-making of teachers.
	- Addressing digital inequalities and promoting digital literacy.
	- Critiquing the capitalist and cognitive exploitation of learners.
	- Redefining the role of technology as a tool for teaching and learning, not a substitute for teachers, schools, or pedagogy.
	- Problem-solving and critical thinking as the core of teaching and learning.
	- Resignifying the role of schools as spaces for problematizing reality.
	- Redimensioning the role of teachers as facilitators of learning, not just transmitters of knowledge.

Note: Due to the amount of data, AI Assist had to limit the analysis to 26 segments from 1 documents.

#### **4. Impact of artificial intelligence on academic research**

In the academic panorama, artificial intelligence (AI) thus not only represents a technological advancement that will facilitate processes but also acts as a catalyst propelling academic research forward. The integration of AI into educational research will enable the transcendence of boundaries, moving from a few textual data to massive corpora, thereby enhancing collaborative research. Consequently, the impact of AI on research methodology cannot be underestimated. Although the robustness of quantitative and qualitative approaches with software was already significant, patterns that the human eye cannot detect immediately will be discerned by AI vision.

However, the integration of AI into academia is not without challenges. Ethical considerations emerge to the forefront, particularly regarding data privacy, bias, and authenticity. A mere test of how ChatGPT or Claude invents non-existent sources for citation reveals that their current algorithms are not immune to bias and could fall into pseudo-analysis. Therefore, the ethical landscape also requires development to ensure that the use of AI in research adheres to clear guidelines and strict ethical principles.

In academia, AI can also support the regulation of publications by streamlining peer analysis, identifying potential reviewers according to their publication history, and classifying similarities through modeling of themes [8], as well as accelerating plagiarism detection. Perhaps in the future, it will evaluate the validity and reliability of findings, improving publication processes and generating new standards of quality and authenticity.

While the capabilities of platforms integrated with AI, such as MAXQDA or ATLAS.ti, in data analysis are fundamental, the impact of AI in academia is expansive, permeating various facets of research, ethics, pedagogy, and publication. Navigating through the opportunities and challenges presented by AI necessitates a balanced approach. The future of academia, underpinned by AI, heralds an era of enhanced research capabilities, data-driven pedagogical strategies, and streamlined publication processes, propelling academic research towards a future that is not only technologically advanced but also ethically aware and methodologically robust.

#### **5. Conclusions**

Although [12] asserts that the innovative approaches brought forth by artificial intelligence liberate research from human bias, thereby enhancing the reliability and validity of investigations, it is imperative to acknowledge that entrusting total responsibility to artificial intelligence and its artificial neural network is not yet feasible. This is because artificial intelligence does not obviate the necessity for researchers to engage in reflective thinking to attain a profound understanding of the texts analyzed. Consequently, AI does not supplant humans in research; rather, it transforms into a supportive tool designed to enhance and expedite the processes of data analysis and systematization, ensuring a meticulous and efficient exploration of information.

Undoubtedly, the emerging trends in research within the social sciences and education, as articulated by Kariri [18] will necessitate processes wherein information

#### *Artificial Intelligence in Educational Research DOI: http://dx.doi.org/10.5772/intechopen.113844*

analysis employs machine and deep learning. This is already exemplified in sentiment analysis [29] and text classification, where these advanced computational techniques facilitate a more nuanced and efficient exploration of data, thereby enriching the analytical depth and breadth of scholarly investigations in these domains. The exponential growth of textual information in the academic sphere necessitates a paradigm shift towards more efficient and rapid data analysis methodologies. Text mining, underpinned by artificial intelligence and Natural Language Processing, emerges as a transformative tool, particularly in the realm of social sciences and education. This is especially pertinent in educational contexts where qualitative elements often outweigh quantitative ones. The study focuses on analyzing texts generated by key stakeholders in education, such as students and educators, to explore the transformative potential of text mining in qualitative methodologies and to pave new avenues for text-based qualitative research.

Text mining has evolved to become an accessible tool for researchers, obviating the need for programming expertise. The advent of AI-integrated qualitative data analysis software like ATLAS.ti and MAXQDA, particularly since the beginning of 2023, has significantly expedited the coding and analysis processes.

The integration of deep learning models and neural networks into text mining portends a future wherein even more complex patterns can be discerned from large data sets. However, it is also crucial to acknowledge that the ability to recognize the context of sentences or documents [30] remains a limitation that must be overcome, considering regional dialects that can lead to varying meanings and, consequently, to interpretative biases. Nevertheless, it is vital to recognize that what was once unattainable in traditional qualitative research, namely the massive analysis of data, is now within reach. These advancements are poised to revolutionize educational interventions; coupled with findings from empirical research, they will contribute to a more nuanced understanding of complex educational processes, thereby enhancing the depth and efficacy of scholarly inquiries.

AI, bolstered by NLP, will facilitate the analysis of large volumes of textual data in both educational and social science contexts, thereby enhancing the outcomes of qualitative research in unprecedented ways. The key advantages include:


While AI-assisted text mining offers numerous advantages, it also poses challenges, such as the potential to overlook the need for human oversight and the urgent requirement for clear ethical guidelines in scientific production. Future research should focus on optimizing these tools for more accurate and nuanced analyses, thereby ensuring the responsible and effective utilization of AI in academic research.

### **Conflict of interest**

"The author declares no conflict of interest."

### **Author details**

Ulises Alejandro Duarte Velazquez Universidad Vasco de Quiroga, Morelia, Michoacán, México

\*Address all correspondence to: aduartev@uvaq.edu.mx

© 2024 The Author(s). Licensee IntechOpen. This chapter is distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/3.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

*Artificial Intelligence in Educational Research DOI: http://dx.doi.org/10.5772/intechopen.113844*

#### **References**

[1] Anawis M. Text mining: The next data frontier. Research & Development World. 2014. Available from: https://www.rdworldonline.com/ text-mining-the-next-data-frontier/

[2] Latard B, Weber J, Forestier G, Hassenforder M. Towards a semantic search engine for scientific articles. Research and Advanced Technology for Digital Libraries. In: 21st International Conference on Theory and Practice of Digital Libraries, TPDL 2017, Thessaloniki, Greece, September 18-21, 2017, Proceedings. Vol. 10450. 2017. pp. 608-611. DOI: 10.1007/978-3-319-67008-9\_54

[3] Mysore S, Jasim M, Song H, Akbar S, Randall AKC, Mahyar N. How data scientists review the scholarly literature. In: Gwizdka J, Rieh SY, editors. CHIIR '23 Proceedings of the 2023 Conference on Human Information Interaction and Retrieval. 2023. pp. 137- 152. DOI: 10.1145/3576840. ISBN: 9798400700354

[4] Stein-Sparvieri E. Text mining e inferencia de defensas en el análisis del discurso en psicología. Subjetividad y Procesos Cognitivos. 2010;**14**(2):304-313

[5] Zanini N, Dhawan V. Text Mining: An Introduction to Theory and some Applications. Research Matters: A Cambridge Assessment Publication; 2015;**19**:38-44. Available from: https:// www.cambridgeassessment.org.uk/ Images/466185-text-mining-anintroduction-to-theory-and-someapplications-.pdf

[6] Feldman R, Sanger J. Introduction to text mining. In: The Text Mining Handbook: Advanced Approaches in Analyzing Unstructured Data. Cambridge: Cambridge University Press; 2006. pp. 1-18

[7] Maulud DH, Zeebaree SRM, Jacksi K, Sadeeq MAM, Sharif KH. State of art for semantic analysis of natural language processing. Qubahan Academic Journal. 2021;**1**(2):21-28. DOI: 10.48161/QAJ. V1N2A44

[8] Jelodar H, Wang Y, Rabbani M, Xiao G, Zhao R. A collaborative framework based for semantic patients-behavior analysis and highlight topics discovery of alcoholic beverages in online healthcare forums. Journal of Medical Systems. 2020;**44**(5):101. DOI: 10.1007/ s10916-020-01547-0

[9] Aleem A, Gore MM. Educational data mining methods: A survey. In: 2020 IEEE 9th International Conference on Communication Systems and Network Technologies (CSNT). Gwalior, India: IEEE; 2020. pp. 182-188. DOI: 10.1109/CSNT48778.2020.9115734. ISSN: 2329-7182

[10] Grimmer J, Stewart BM. Text as data: The promise and pitfalls of automatic content analysis methods for political texts. Political Analysis. 2013;**21**(3):267- 297. DOI: 10.1093/PAN/MPS028

[11] Munguia P, Brennan A. Scaling the student journey from course-level information to program level progression and graduation: A model. Journal of Learning Analytics. 2020;**7**(2):84-94. DOI: 10.18608/jla.2020.72.5

[12] Bozkurt A, Karadeniz A, Baneres D, Guerrero-Roldán AE, Rodríguez ME. Artificial intelligence and reflections from educational landscape: A review of AI studies in half a century. Sustainability. 2021;**13**(2):800. DOI: 10.3390/su13020800

[13] Lin B, Ghaddar B, Hurst A. Text mining undergraduate engineering

programs' applications: The role of gender, nationality, and socioeconomic status (arXiv:2107.14034; Versión 4). arXiv. 2022. DOI: 10.48550/ arXiv.2107.14034

[14] Roberts ME, Stewart BM, Airoldi EM. A model of text for experimentation in the social sciences. Journal of the American Statistical Association. 2016;**111**(515):988-1003. DOI: 10.1080/01621459.2016.1141684

[15] Cheng M-Y, Kusoemo D, Gosno RA. Text mining-based construction site accident classification using hybrid supervised machine learning. Automation in Construction. 2020;**118**:103265. DOI: 10.1016/j. autcon.2020.103265

[16] Duarte Velazquez UA. Análisis con minería de datos del Marco Curricular y el Plan de Estudios 2022 de la Educación Básica Mexicana. Transdigital. 2022;**3**(6): Article 6. Available from: https://doi. org/10.56162/transdigital122

[17] Tsai Y-C. Learner-centered analysis in educational metaverse environments: Exploring value exchange systems through natural interaction and text mining (arXiv:2305.08326; Versión 1). arXiv. 2023. DOI: 10.48550/ arXiv.2305.08326

[18] Kariri E, Louati H, Louati A, Masmoudi F. Exploring the advancements and future research directions of artificial neural networks: A text mining approach. Applied Sciences. 2023;**13**(5):3186. DOI: 10.3390/ app13053186

[19] Holzinger A. Interactive machine learning for health informatics: When do we need the human-in-the-loop? Brain Informatics. 2016;**3**(2):119-131. DOI: 10.1007/s40708-016-0042-6. PMID: 27747607; PMCID: PMC4883171

[20] Young T, Hazarika D, Poria S, Cambria E. Recent trends in deep learning based natural language processing. IEEE Computational Intelligence Magazine. 2018;**13**(3):55-75

[21] Dey I. Qualitative Data Analysis: A User-Friendly Guide for Social Scientists. New York: Routledge; 1993. DOI: 10.4135/9781483384436

[22] Guest G, MacQueen KM, Namey EE. Applied Thematic Analysis. SAGE Publications, Inc.; 2012. DOI: 10.4135/9781483384436

[23] Mayring P. Qualitative content analysis: Theoretical background and procedures. In: Bikner-Ahsbahs A, Knipping C, Presmeg N, editors. Approaches to Qualitative Research in Mathematics Education: Examples of Methodology and Methods. Netherlands: Springer; 2015. pp. 365-380. DOI: 10.1007/978-94-017-9181-6\_13

[24] Schreier M. Qualitative Content Analysis in Practice. London: SAGE Publications Ltd; 2012

[25] Kuckartz U. Discipline: Anthropology methods: Textual analysis, secondary data analysis, qualitative data analysis. In: Qualitative Text Analysis: A Guide to Methods, Practice and Using Software. SAGE Publications Ltd; 2014. DOI: 10.4135/9781446288719. ISBN: 9781446288719

[26] Muhr T. ATLAS/ti—A prototype for the support of text interpretation. Qualitative Sociology. 1991;**14**(4):349- 371. DOI: 10.1007/BF00989645

[27] ATLAS.ti. ATLAS.ti Windows. Analyze Anything, faster. Germany: ATLAS.ti Scientific Software Development GmbH; 2023. Available from: https://doc.atlasti.com/ QuicktourWin.es/index.html

*Artificial Intelligence in Educational Research DOI: http://dx.doi.org/10.5772/intechopen.113844*

[28] Kuckartz U. Textanalyse mit QDA— Programmen. In: Kuckartz U, editor. Computergestützte Analyse qualitativer Daten: Eine Einführung in Methoden und Arbeitstechniken. VS Verlag für Sozialwissenschaften. Opladen: Westdt; 1999. pp. 15-46. DOI: 10.1007/978-3-322- 86592-2. ISBN-13: 978-3-531-22178-6 e-ISBN-13: 978-3-322-86592-2

[29] Zhang Y, Wallace B. A sensitivity analysis of (and practitioners' guide to) convolutional neural networks for sentence classification. 2015. Available from: arXiv.Org. https://arxiv.org/ abs/1510.03820v4

[30] Widiastuti NI. Deep learning— Now and next in text mining and natural language processing. IOP Conference Series: Materials Science and Engineering. 2018;**407**(1):012114. DOI: 10.1088/1757-899X/407/1/012114

#### **Chapter 3**

### Recognition of Brain Wave Related to the Episode Memory by Deep Learning Methods

*Takashi Kuremoto, Junko Ishikawa, Shingo Mabu and Dai Mitsushima*

#### **Abstract**

Hippocampus makes an important role of memory in the brain. In this chapter, a study of brain wave recognition using deep learning methods is introduced. The purpose of the study is to match the ripple-firings of the hippocampal activity to the episodic memories. In fact, brain spike signals of rats (300–10 kHz) were recorded and machine learning methods such as Convolutional Neural Networks (CNN), Support Vector Machine (SVM), a deep learning model VGG16, and combination models composed by CNN with SVM and VGG16 with SVM were adopted to be classifiers of the brain wave signals. Four kinds of episodic memories, that is, a male rat contacted with a female/male rat, contacted with a novel object, and an experience of restrain stress, were detected corresponding to the ripple waves of Multiple-Unit Activities (MUAs) of hippocampal CA1 neurons in male rats in the experiments. The experiment results showed the possibility of matching of ripple-like firing patterns of hippocampus to episodic memory activities of rats, and it suggests disorders of memory function may be found by the analysis of brain waves.

**Keywords:** brain wave, ripple-like wave, episodic memory, pattern recognition, convolutional neural networks

#### **1. Introduction**

Brain activities and functions can be recorded by Electroencephalograph (EEG). EEG signals have been widely used in clinical examination, psychophysiology, engineering, etc. To realize the Brain Computer Interface (BCI), there have been many studies to recognize mental tasks by EEG signals and machine learning methods in the last decades [1–9]. Meanwhile, EEG electrodes are usually placed along the scalp of human or animals in noninvasive measurement methods, the activities of deep structure in the brain, such as the base of cortical gyrus, hippocampus, thalamus, and brain stem, are not recorded exactly by EEG data. To investigate the relationship between the hippocampus and memory function, Ishikawa, Tomokage, and Mitsushima recorded Multiple-Unit Activities (MUA) of CA1 neurons in adult rats that experienced four kinds of episodes (each for 10 minutes): restraint stress, social interaction

with a female or male rat, or observation of a novel object [10]. According to the signal analysis of MUA, they hypothesized that early learning process [11] exists in CA1 pyramidal neurons and experience-induced super bursts, synaptic plasticity, and ripple-like firings. Sharp waves and ripples, which can be observed in the brain waves of CA1 and CA3 neurons of hippocampus with 140–300 Hz, are considered to be concerned with memory consolidation [12–15].

In this chapter, machine learning methods such as Support Vector Machines (SVM) [16], Convolutional Neural Networks (CNN) [17], VGG16 (a well-known deep CNN with 16 layers [18, 19]), and hybrid models which are CNN with SVM [17], and VGG16 with SVM are introduced to recognize the MUA patterns related to episodic memories.

#### **2. Multiple-Unit Activities (MUA) of CA1 neurons**

Signal data of Multiple-Unit Activities (MUA) of CA1 neurons in hippocampus were recorded by vertically movable recording electrodes (Unique Medical Co., LTD, Japan), which were implanted in adult Sprague-Dawley rats (CLEA Japan Inc., Japan) at age 15 to 25 weeks in Yamaguchi University, Japan [10]. Neural signals were measured with a shielded cable and amplifiers. We used 300–10,000 Hz band-pass filter to obtain original signals of MUA. The sampling rate of the band-pass filtered signals was 25 kHz for the digital time series data by a CED 1401 interface (Cambridge Electronics Design, U.K.). The environment of the rat and the electrode used for MUA recording are shown in **Figure 1**.

Experient events for episodic memories of male rats performed in the experiment were restraint stress, social interaction with a female or male rat, or observation of a novel object which was a yellow plastic brick (LEGO bricks) [10]. The schedule of MUA recording, including preparation (15 minutes), experience (10 minutes), and consolidation (30 minutes), is shown in **Figure 2**.

Four kinds of original MUA signals related to the different experiences are shown in **Figure 3**. The duration of each signal is 1 second, that is, 25,000 time series data. Notice the scale of the vertical axis (amplitude) is different in these signals.

An example of no ripple MUA signals is depicted in **Figure 4**. It was also adopted in our last classification experiment as a kind of input data.

**Figure 1.**

In vivo *recording of MUA [10]. (a) An adult male rat implanted with electrodes. (b) Electrode used for recording MUA.*

*Recognition of Brain Wave Related to the Episode Memory by Deep Learning Methods DOI: http://dx.doi.org/10.5772/intechopen.112531*

**Figure 2.**

*Schedule of MUA recording and data used in classification experiment [16].*

**Figure 3.**

*Four kinds of patterns of MUA signals related to different experiences (horizontal axis: time (1/25,000sec), vertical axis: volt). (a) Restraint stress (b) With a female rat (c) With a male rat (d) With an object.*

**Figure 4.** *An example of no ripple MUA signals (horizontal axis: time (1/25,000sec), vertical axis: volt).*

#### **3. Machine learning methods**

To analyze brain waves, various statistical methods and machine learning approaches have been suggested since 1950s [1, 4–9]. To classify different EEG signals, methods such as Principal Component Analysis (PCA), Artificial Neural Networks (ANN), Support Vector Machines (SVM), and Deep Learning (DL), have been widely utilized. To recognize ripple-like waves of MUA of the hippocampus related to episodic memories, we confirmed the performance of SVM [16], Convolutional Neural Networks (CNN) [17], VGG16 [18, 19], and hybrid models such as CNN with SVM [17], and VGG16 with SVM by the experiment using the MUA data described in the last section.

#### **3.1 SVM**

Support Vector Machines (SVM) [16] showed a higher performance for pattern recognition problems than other classifiers such as Multi-Layered Perceptron (MLP), which is a kind of feedforward neural network and led to the second Artificial Intelligence (AI) boom in the middle of 1980s. For a two-class classification problem, the decision function of a linear SVM is as follows.

$$f(\mathbf{x}) = \text{sign}\left(\sum\_{i} v\_i(\mathbf{x} \bullet \mathbf{x}\_i) + b\right) \tag{1}$$

where *x* is the unknown input data (e.g., a high-dimension vector), *xi* is so-called a support vector i, chosen from training data (teacher signals) which determines the hyperplanes with parameters *vi* and *b*. Parameters in (1) are obtained with training data and Quadratic Programming (QP).

A nonlinear SVM maps the training data into a higher-dimensional feature space with a nonlinear kernel function *k x*, *y* � �, then classifies the data by a separating hyperplane with maximum margin in the feature space.

*Recognition of Brain Wave Related to the Episode Memory by Deep Learning Methods DOI: http://dx.doi.org/10.5772/intechopen.112531*

$$f(\mathbf{x}) = \text{sign}\left(\sum\_{i} \nu\_i k(\mathbf{x}, \mathbf{x}\_i) + b\right) \tag{2}$$

$$k(\mathbf{x}, \mathbf{x}\_i) = \exp\left(-\left\|\mathbf{x} - \mathbf{x}\_i\right\|^2 / \left(2\sigma^2\right)\right) \tag{3}$$

For a multi-class problem, a one-vs-the-rest SVM algorithm needs to be adopted, that is, for a *m*-class problem, *m* SVMs are built at first, then the class of an unknown data is voted by these SVMs.

In the pattern recognition problem, four kinds of MUA signals were applied by SVM, 50.0% accuracy was obtained when 3,600 training data (900 for each kind of MUA signal) were utilized and each data with 25,000 dimensions (a signal in one second with a sampling rate 25.0 kHz).

#### **3.2 CNNs**

Convolutional Neural Networks (CNN) are mostly adopted in deep learning methods. For the MUA pattern recognition in this study, two shallow models of CNN, CNN-1 and CNN-2, were used as shown in **Figure 5** [17] and **Figure 6**. CNN-2 was structured by more convolutional layers and pooling layers for the investigation of the change in recognition accuracy. The recognition experiment using the four kinds of MUA signals, as shown in **Figure 2**, showed that the deeper model CNN-2 had a higher performance than CNN-1. The identification rates (validation accuracies) of CNN-1and CNN-2 were 65.78 and 86.09%, respectively, when 1,328 signals of MUA were utilized to train CNNs, detailly, the number of different experiences was restraint stress 298, with a female rat 306, with a male rat 330, with a novel object 270, respectively. So, it is necessary to investigate the performance of more deeper CNN models for this pattern recognition problem and we tested the cases of VGG16 [18–20] and ResNet50 [21].

**Figure 5.**

*CNN1: a shallow neural network for MUA pattern recognition [17].*

**Figure 6.**

*CNN-2: an improved neural network for MUA pattern recognition.*

#### **3.3 VGG and ResNet**

A deep learning model VGG16 [18–20] is well-known for its high performance in ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, which is an image pattern recognition contest since 2010. VGG16 was trained by 1,000 classes of 14,000,000 images, and its error rate was 7.3%, meanwhile, the champion's score of ILSVRC 2012 was 15.3%. The structure of VGG16 was shown in **Figure 7** [20].

**Figure 7.** *The structure of VGG16 [20].*

*Recognition of Brain Wave Related to the Episode Memory by Deep Learning Methods DOI: http://dx.doi.org/10.5772/intechopen.112531*

Convolutional layers and max-pooling layers and fully connected layers with ReLU units were composed repeatedly in VGG16.

A more complex deep learning model Residual Network (ResNet) won the champion of ILSVRC 2015 with an error rate of 3.57% (top-5 validation error). The comparison of structures between a conventional deep learning model VGG19 and a ResNet with 34 layers is shown in **Figure 8** [21].

**Figure 8.**

*The structures of VGG19 and ResNet34 [21].*

ResNet solved the problem that deeper networks resulted in training error degradation problem by introducing a deep residual learning framework, that is, "shortcut connections" were added to the conventional plain networks.

In our recognition experiment, five kinds of MUA signals as shown in **Figures 3** and **4** were preprocessed by Short-Time Fourier Transform (STFT), which resulted in higher recognition rates than by inputting the original values of signals to classifiers. The size of window function of STFT was 1,024, and the overlap of shift of windows was 256 for a signal with 25,000 data. A sample of the preprocessed data is shown in **Figure 9**. In fact, we compared the recognition rates between the case of original MUA signals and the case of STFT data using a VGG16 model without fine-tuning, and they were 70.92, 74.04% (val\_acc), respectively.

The learning performance of CNN-2, VGG16, and ResNet50 using the power spectrum data is shown in **Figure 10**. As a result, the validation accuracies of these classifiers were 88.95, 93.57, and 62.47%. It suggested that too shallow or deep structures of deep learning models were not suitable for the 5-class recognition problem of MUA signals, and VGG16 was the best of them. Note that the horizontal axes in **Figure 10** were the number of training iterations, and it was 1,000 in the case of CNN-2, 200 in the cases of VGG16 and ResNet50. The reason for these differences is that the satisfied training needs to be finished when the training accuracy is converged.

All training accuracies and training losses of CNN-2, VGG16, ResNet50 were converged, however, the validation accuracy of ResNet50 vibrated intensely. The reason can be thought that the model scale was too large for the problem, or in other words, the number of training samples and classes of MUA signals was too few for the optimization of a robust ResNet50.

#### **3.4 Hybrid models**

Generally, deep convolutional neural networks have fully connected layers, which are the classical Multi-Layered Perceptron (MLP). As Support Vector Machine (SVM) showed its priority to MLP for classification problems, we proposed a kind of hybrid model by adopting SVM to the deep learning models instead of MLP in our previous works [9, 17]. The hybrid models are composed of deep convolutional neural networks such as CNN-1, CNN-2, VGG16, which are optimized by

**Figure 9.** *A power spectrum data of a MUA signal.*

*Recognition of Brain Wave Related to the Episode Memory by Deep Learning Methods DOI: http://dx.doi.org/10.5772/intechopen.112531*

**Figure 10.** *Learning curves of different deep learning models. (a) CNN-2 (b) VGG16 (c) ResNet50.*

#### **Figure 11.**

*Comparison of validation accuracies of different deep learning models.*

transfer-learning (fine-tuning) with training data and replacing the output layers (such as full-connected layers and SoftMax output function) by SVMs. Using the 5-class data descripted in Section 3.2, we investigated the recognition rates of the conventional deep convolutional neural networks CNN-2 and VGG16 and the hybrid models CNN-2 with SVM and VGG16 with SVM by 5-fold cross-validation experiments. As shown in **Figure 11**, two hybrid models showed their priorities, and the validation accuracy of VGG16 with SVM even reached 95.60% for the 5-class problem of MUA signals.

The classification results of the validation data were plotted by *t*-SNE method [22] as shown in **Figure 12**. It can be confirmed that VGG16 with SVM which had the highest recognition rate as shown in **Figure 11** separated the five kinds of MUA signals better than CNN-2 with SVM.

#### **4. Ripple-like waves related to episodic memories**

In last section, the MUA signals of rat's hippocampus CA1 neurons corresponding to different episodic memories were classified by kinds of deep learning methods. It is interesting to discover the difference between these signals, especially, the pattern of different ripple-like waves among them. Features of input data can be extracted by deep learning models by convolutional layers or pooling layers and they can be expressed by visual explanations: Gradient-weighted Class Activation Mapping (Grad-CAM) [23].

In **Figure 13**, an original MUA signal (above) and its features (under) in thermography expression given by Grad-CAM were shown. High temperatures mean high participation degrees of the features. In this case, 41 features which were the convolutional layer of CNN-1 were evaluated for the original input image (125,000), that is, 5,000 data of the signal corresponding to one feature, and the convolution window of the signal slid 500 of the time series data. The most important feature for classification by CNN-1 was No.24 in red color, corresponding to the data from 11,500 to 16,500 as shown in the above of **Figure 13**. This duration of the MUA signal suggested a kind of ripple-like wave pattern concerning to episodic memory.

*Recognition of Brain Wave Related to the Episode Memory by Deep Learning Methods DOI: http://dx.doi.org/10.5772/intechopen.112531*

#### **Figure 12.**

*Visual comparison of labeled validation data given by CNN-2 with SVM and VGG16 with SVM according to* t*-SNE method [22]. Letters* r*,* f*,* m*,* o*,* n *denote MUA signals of* restrain stress*,* with a female rat*,* with a male rat*,* with an object*, and* no ripples*. (a) CNN-2 with SVM (b) VGG16 with SVM.*

In Ref. [17], four kinds of MUA signals that are related to a rat's experiences of *restrain stress*, *with a female rat*, *with a male rat*, *with an object*, and *no ripples* were analyzed by CNN-1 and Grad-CAM. Different from the case in **Figure 13**, in **Figure 14**

**Figure 13.**

*The relationship between an original MUA signal and its 41 features given be Grad-CAM [23].*

[17], there were 2,500 features were expressed for slide number 10, while the length of the MUA signal was 25,000 (1sec, 25kHz). Duration of episodic memory-related signals were extracted by choosing the highest temperatures in heatmaps as shown in **Figure 15**.

To enclose the distribution of the frequencies of ripple-like waves, Fourier transform and Cepstrum transform were performed to the durations of signals shown in **Figure 15**. The highest spectra of these duration point to the frequencies of 100–400 Hz are as shown in **Figure 16**.

Cepstrum transform, which is another signal analysis method, estimates periodic structures in frequency spectra by the inverse Fourier transform of the logarithm of the Fourier transform results. The Cepstrum transform results of the four kinds of MUA signals are shown in **Figure 17**. Compare to the Fourier transform result, it can be more easy to find that the patterns of *restrain stress* in **Figure 17(a)** and *with a male rat* **Figure 17(b)** are similar to each other, meanwhile, ripple-like waves of *a female rat* **Figure 17(c)** and *with an object* **Figure 17(d)** are similar.

To verify the data distributions of Fourier transform results and Cepstrum transform results, Principal Component Analysis (PCA) was adopted and the 2-dimensional results of PCA are shown in **Figure 18**. The similarities of the four MUA signals are more easy to be confirmed in the case of Cepstrum transform.

#### **5. Conclusions**

It is well-known that Sharp Waves and Ripples (SWR) in brain waves are related to memory consolidation. In this chapter, to classify Multiple-Unit Activities (MUA) of hippocampus CA1 neurons, machine learning methods, such as Support Vector Machine (SVM), Convolutional Neural Networks (CNNs), deep learning models such as VGG16 and hybrid models CNN with SVM and VGG16 with SVM were introduced. The MUA data of rats were recorded by movable electrodes implanted above the hippocampal CA1 and fixed with dental cement. Four kinds of events were

*Recognition of Brain Wave Related to the Episode Memory by Deep Learning Methods DOI: http://dx.doi.org/10.5772/intechopen.112531*

**Figure 14.**

*MUA signals (above) and their heatmaps (under) of the event concerned features in CNN-1 [17] (horizontal axis in blue heatmap: feature number). (a) Restraint stress (b) With a male rat (c) With a female rat (d) With an object.*

#### **Figure 15.**

*Ripple-like waves related to the different episodic memories (Scale of horizontal axis: 1/25,000 sec; vertical axis: volt) [17]. (a)Restraint stress (b) With a male rat (c) With a female rat (d) With an object.*

experienced by a male adult rat: *restraint stress*, *with an intruder male rat*, *with a female rat*, and *with a novel objec*t (a yellow LEGO® brick). The highest recognition rate (validation accuracy) of the four kinds of ripple-like waves and a no-ripple signal was 95.6% given by VGG16 with SVM method. The duration of ripple-like waves of the original MUA signals were extracted by Grad-CAM, and spectra analysis of the relationship between these episodic memory-related signals were performed by Fourier transform, Cepstrum transform, and Principal Component Analysis (PCA).

Although the MUA data used in this experiment were recorded from male rats, the pattern of ripple firing associated with a particular episodic experience may be common to various animal species, including humans [24]. If commonality exists, it may be possible to identify the neural circuits that generate those patterns and extract experiential information from different animal species. These questions need to be

*Recognition of Brain Wave Related to the Episode Memory by Deep Learning Methods DOI: http://dx.doi.org/10.5772/intechopen.112531*

#### **Figure 16.**

*Fourier transform result of the specified intervals of MUA extracted by Grad-CAM (Scale of horizontal axis: Hz) [17]. (a) Restraint stress (b) With a male rat (c) With a female rat (d) With an object.*

**Figure 17.**

*Cepstrum transform results of the important intervals of MUA signals extracted by Grad-CAM (Scale of horizontal axis: msec) [17]. (a) Restraint stress (b) With a male rat (c) With a female rat (d) With a novel object.*

*Recognition of Brain Wave Related to the Episode Memory by Deep Learning Methods DOI: http://dx.doi.org/10.5772/intechopen.112531*

**Figure 18.**

*Data distribution of different transforms of ripple-like waves by PCA [17]. (a) The case of Fourier transform. (b) The case of Cepstrum transform.*

answered in order to decipher memory information, and their resolution has the potential to open up a new era in brain sciences.

#### **Acknowledgements**

We would like to thank Mr. Takaaki SASAKI, Mr. Yuichi KOBAYSHI for them hard work involved with the recognition experiments of this study.

The study was supported JSPS KAKENHI Grant No. 19H03402, No.20K07724, No.22K12152, No.22H03709.

### **Conflict of interest**

The authors declare no conflict of interest.

### **Author details**

Takashi Kuremoto<sup>1</sup> \*, Junko Ishikawa<sup>2</sup> , Shingo Mabu2 and Dai Mitsushima<sup>2</sup> \*

1 Institute of Technology, Saitama, Japan

2 Yamaguchi University, Yamaguchi, Japan

\*Address all correspondence to: kuremoto.takashi@nit.ac.jp and mitsu@yamaguchi-u.ac.jp

© 2023 The Author(s). Licensee IntechOpen. This chapter is distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/3.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

*Recognition of Brain Wave Related to the Episode Memory by Deep Learning Methods DOI: http://dx.doi.org/10.5772/intechopen.112531*

#### **References**

[1] Blankertz B et al. The BCI Competition 2003: Progress and perspectives in detection and discrimination of EEG Single Trials. IEEE Transaction on Biomedical Engineering. 2004;**51**(6):1044-1051

[2] Colorado State University, Brain-Computer Interfaces Laboratory. Available from: http://www.cs.colostate. edu/eeg/

[3] BCI competition II. Available from: http://www.bbci.de/competition/ii/ #datasets

[4] Chin ZY, Ang KK, Wang C, Guan C, Zhang H. Multi-class filter bank common spatial pattern for four-class motor imagery BCI. In: Annual International Conference of the IEEE Engineering in Medicine and Biology Society, 2009. Vol. 2009. Minneapolis, MN, USA: EMBC; 2009. pp. 571-574

[5] Tang Z, Li C, Sun S. Single-trial EEG classification of motor imagery using deep convolutional neural networks. Optik. 2017;**130**:11-18

[6] Schirrmeister RT, et al. Deep learning with convolutional neural networks for EEG decoding and visualization, arXiv: 1703.05051v5 [cs.LG]. 2018

[7] Kuremoto T, Baba Y, Obayashi M, Mabu S, Kobayashi K. Enhancing EEG signals recognition using ROC curve. Journal of Robotics, Networking and Artificial Life. 2018;**4**(4):283-286

[8] Kuremoto T, Baba Y, Obayashi M, Mabu S, Kobayashi K. Mental task recognition by EEG signals: A novel approach with ROC analysis. In: Anbarjafari G, Escalera S, editors. Chapter 4Human-Robot Interaction–Theory and Application. InTech; 2018. pp. 65-78

[9] Kuremoto T, Sasaki T, Mabu S. Mental task recognition using EEG signal and deep learning methods. Stress Brain and Behavior. 2019;**1**:18-23

[10] Ishikawa J, Tomokage T, Mitsushima D. A possible coding for experience: Ripple-like events and synaptic diversity. BioRxiv. 2019:1-45. DOI: 10.1101/2019.12.30.891259

[11] Mitsushima D, Sano A, Takahashi T. A cholinergic trigger drives learninginduced plasticity at hippocampal synapses. Nature Communications. 2013;**4**:2760. DOI: 10.1038/ncomms3760

[12] Buzsaki G. Hippocampal sharp wave-ripple: A cognitive biomarker for episodic memory and planning. Hippocampus. 2015;**25**:1073-1188

[13] Joo HR, Frank LM. The hippocampal sharp wave-ripple in memory retrieval for immediate use and consolidation. Nature Reviews Neuroscience. 2018;**19**:744-757

[14] Kay K, Frank LM. Three brain states in the hippocampus and cortex. Hippocampus. 2019;**29**:184-238

[15] Fenandez-Ruiz A et al. Longduration hippocampal sharp wave ripples improve memory. Science. 2019; **264**:1082-1086

[16] Hearst MA et al. Support vector machines. IEEE Intelligent Systems; July/August 1998

[17] Kuremoto T, Ishikawa J, Sasaki T, Mabu S, Mitsushima D. Matching the ripple-wave to the episodic memory – a case study of rat. Stress Brain and Behavior. 2023;**1**(e02204):19-30

[18] Russakovsky O, Deng J, Su H, et al. ImageNet large scale visual recognition

challenge. International Journal of Computer Vision. 2015;**115**(3):211-252

[19] Simonyan K, Zisserman A. Very deep convolutional networks for largescale image recognition. 2014. arXiv: 1409.1556

[20] New Technology Life Style. Available from: https://newtechnolog ylifestyle.net/vgg16originalpicture/. 2019

[21] He K, Zhang X, Ren S, Sun J. Deep residual Learning for image recognition. arXiv:1512.03385v1. 2015

[22] Maaten L, Hinton G. Visualizing data using t-SNE. Journal of Machine Learning Research. 2008;**9**:2579-2606

[23] Sevaraiu RR, et al. Grad-CAM: Explanations from deep networks via gradient-based localization. arXiv: 1610.02391v4 [cs.CV]. 2019

[24] Buzsáki G, Logothetis N, Singer W. Scaling brain size, keeping timing: evolutionary preservation of brain rhythms. Neuron. 2013;**80**(3):751-764

Section 2
