**4. Evaluation experiment**

In the survey report on study trend published in 2004, Tatsubori et al. used the publication databases of IBM to survey the research trend of software architecture after 1999. They manually extracted 51 major papers [8]. In this study, we examine how well the automated method can extract the major papers that were manually extracted by experts.

Academic Landscape Based on Network Analysis

as major papers.

**5. Conclusion** 

**6. References** 

et al. for quantitative evaluation of research trend.

Fig. 6. An image of extraction of papers by SciHi

task we will examine these papers to see what meaning they have.

Considering Analysis of Variation in the Years of Lucubration Publishing 377

30, 32~39, 41~43 and 49~51). Regarding five roles in Fig.5., we manually made minute adjustment for SciHi (see Fig.6.) because they were an original classification set by Tatsubori

As shown in Fig.6., visualization result by the page rank algorithm (hereafter, it is referred as this algorithm) with the variance value of a target paper taken into consideration as weight showed all the 23 papers were extracted as major papers. Especially, the papers (44~58 in Fig.6.) were extracted manually by Tatsubori et al. and all of them were extracted

Thus, this algorithm can automatically extract major papers that are the same papers manually extracted by experts. However, SciHi extracted major papers that were not extracted by Tatsubori et al. such as a large (green) node in the cluster of [R]. As a future

In this study on the network analysis of quoted academic papers, we tried automated extraction of major papers in each one of the fields identified by clustering. Specifically, we analyzed the variance of the release years of papers quoting the target paper; applied the result to the page rank algorithm to calculate the importance of the target paper. Then, we examined how well the automated method can extract the major papers that were extracted by experts by comparing the result with the report on research trend in the software architecture field that was published by Tatsubori et al. in 2004. As a result, this algorithm could extract all the target papers as major papers. In other words, this algorithm could automatically extract major papers with the same result as what experts manually obtained. Last of all, in our experiment for evaluation, SciHi also extracted major papers that experts

did not manually. As a future task we will examine what these papers mean.

[1] R・J・Wilson .(2001). *Introduction to Graph Theory*, Kindai Kagaku Sha Co.,Ltd.

*networks*, Physical Review E, Vol. 69.

[2] M. E. J. Newman and M. Girvan.(2004). *Finding and evaluating community structure in* 

Tatsubori et al. used GoogleScholar to obtain the number of quotations for each year and extracted top 40 papers in a year. In addition to them, they extracted 11 more papers about software architecture by identifying international conferences that they thought particularly important, resulting in 51 major papers. Furthermore, they adopted a unique classification method in order to evaluate study trend quantitatively. Specifically, they classified 51 papers by defining each paper focused on which one of the following five roles that software architecture played in software development process. Fig.5. shows the classification result.


Fig. 5. An image of classification of papers by five roles of software architecture

Fig.6. shows the major papers extracted by SciHi in a similar manner to that of Tachibori et al. Fig.5. and Fig.6. both indicate the paper numbers.

The following describes the way of extraction by SciHi: We used SCOPUS as publication databases. Then, we selected "Software Architecture" for query and "1999 to 2004" for the period for extraction of papers. Among 51 papers, 28 papers were not included because they did not exist in SCOPUS (see the papers numbers in Fig.5. : 1, 5, 12~14, 16~19, 21, 24, 27, 29,

30, 32~39, 41~43 and 49~51). Regarding five roles in Fig.5., we manually made minute adjustment for SciHi (see Fig.6.) because they were an original classification set by Tatsubori et al. for quantitative evaluation of research trend.

As shown in Fig.6., visualization result by the page rank algorithm (hereafter, it is referred as this algorithm) with the variance value of a target paper taken into consideration as weight showed all the 23 papers were extracted as major papers. Especially, the papers (44~58 in Fig.6.) were extracted manually by Tatsubori et al. and all of them were extracted as major papers.

Thus, this algorithm can automatically extract major papers that are the same papers manually extracted by experts. However, SciHi extracted major papers that were not extracted by Tatsubori et al. such as a large (green) node in the cluster of [R]. As a future task we will examine these papers to see what meaning they have.

Fig. 6. An image of extraction of papers by SciHi
