**5. Case study for the formation process of research areas**

In this map, we try to understand the formation processes of several research areas through chronological changes of network structure. This section describes two cases for the Internet of Things (IoT) and Brain-Computer Interface (BCI).

In **Figure 10** shows the analytic views for an IoT area from 2012 to 2016, which includes 574 nodes as of 2016. The last view is the analytic view in 2016 displaying >0.6 cosine similarities as edges.

In 2012, four islands (places, at which nodes are densely located) mainly for IoT frameworks and networks and for IoT system and security are barely found (labels of each island have been extracted by the summarization function of feature phrases).

**Figure 10.** Formation of IoT areas.

In 2013, a funding project (orange node) was firstly established in the security, and then the corresponding island grew bigger, that is, the number of articles increased, although a causal relationship is unclear.

We confirmed several other processes of research area formation in our case studies. For example, in **Figure 11** shows the analytic view for BCI in 2016. In this figure, an island at the top is growing while citing articles for several specific research themes, such as medical applications, brain waves, pattern recognition, and steady state visual evoked potentials (SSVEP). Thus, we can understand that the BCI has been simultaneously approached from several different conventional research themes, and is integrating them. In this manner, we confirmed that the

Mapping Science Based on Research Content Similarity http://dx.doi.org/10.5772/intechopen.77067 191

In this study, we developed a map of science, Mapping Science based on the research content similarity for funding project descriptions and recently published articles, which have difficulty in applying the citation analysis. After improving the existing paragraph embedding technique with an entropy-based clustering method of word vectors, we confirmed the good face validity. Then, we introduced the map constructed from approx. 300 k IEEE articles and NSF projects from 2012 to 2016 with the clustering and layout method of articles/projects and analytic functions provided on the map. Finally, we confirmed that formation processes of

formation processes of research areas can be captured by closely observing the map.

some specific research areas can be captured as changes of network structure.

**6. Conclusion and future work**

**Figure 11.** Formation of BCI.

Then, in 2014, the island of the IoT frameworks and networks also had a funding project and grew bigger. At the same time, researchers of each island, which seem to correspond to the different research community, started to recognize with each other, and thus mutual citation links (light green edges) between islands began to be drawn.

In 2015 and 2016, this movement was accelerated; thus, we can confirm that the islands were getting bigger and denser, and mutual citation links increased. Moreover, the other islands than the first four islands, for example, an island for IoT services and semantics at the upperleft corner also gradually grew, and some of them are greatly increasing the articles by getting funding projects.

Finally, the edges of the cosine similarity 0.6 in the last view mean relatively weak similarity described in Section 3.2. In contrast, nodes which compose an island are mutually connected with stronger similarities, although they are too dense to confirm in the figure. Therefore, in this IoT area, there are several research communities dedicated to specific research themes, and they are mutually connected with their content similarity and citation relations. Thus, we can understand that they are developing each theme while forming the IoT area as a whole.

**Figure 11.** Formation of BCI.

In 2013, a funding project (orange node) was firstly established in the security, and then the corresponding island grew bigger, that is, the number of articles increased, although a causal

Then, in 2014, the island of the IoT frameworks and networks also had a funding project and grew bigger. At the same time, researchers of each island, which seem to correspond to the different research community, started to recognize with each other, and thus mutual citation

In 2015 and 2016, this movement was accelerated; thus, we can confirm that the islands were getting bigger and denser, and mutual citation links increased. Moreover, the other islands than the first four islands, for example, an island for IoT services and semantics at the upperleft corner also gradually grew, and some of them are greatly increasing the articles by getting

Finally, the edges of the cosine similarity 0.6 in the last view mean relatively weak similarity described in Section 3.2. In contrast, nodes which compose an island are mutually connected with stronger similarities, although they are too dense to confirm in the figure. Therefore, in this IoT area, there are several research communities dedicated to specific research themes, and they are mutually connected with their content similarity and citation relations. Thus, we can understand that they are developing each theme while form-

links (light green edges) between islands began to be drawn.

relationship is unclear.

**Figure 10.** Formation of IoT areas.

190 Scientometrics

funding projects.

ing the IoT area as a whole.

We confirmed several other processes of research area formation in our case studies. For example, in **Figure 11** shows the analytic view for BCI in 2016. In this figure, an island at the top is growing while citing articles for several specific research themes, such as medical applications, brain waves, pattern recognition, and steady state visual evoked potentials (SSVEP). Thus, we can understand that the BCI has been simultaneously approached from several different conventional research themes, and is integrating them. In this manner, we confirmed that the formation processes of research areas can be captured by closely observing the map.
