**5. Conclusion**

376 New Research on Knowledge Management Models and Methods

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

R: Requirements for the system of various stakeholders are reflected in the

C: If all the requirements cannot be met, the architecture is changed to meet the

S: If the architecture is abstract, it must be modified so that it can be implemented in a working system. Contrary, if the system has some changes, the changes should

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

al. Fig.5. and Fig.6. both indicate the paper numbers.

Fig.6. shows the major papers extracted by SciHi in a similar manner to that of Tachibori et

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,

M: Architect designs the architecture based on the meta-model.

A: The architecture is actually build up/modified.

be reflected to the architecture.

classification result.

architecture.

requirements.

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.
