Academic Evaluation System Based On Graph-theoretical Clustering

The impact factor evaluation of a particular author is being hot debated in nowadaysacademic area. Among those academic impact factors, h-index is an excellent evaluationstandard. By researching on some applications of h-index, several prevalent problems arefound, which are summarized as follows. Firstly, without the support of huge amount ofdata the calculated h index is lacking of reference value. Secondly, the bad distinction ofauthors sharing the same name results in the bigger calculated h-index. Additionally, overmatching approach and some of the excessive self-reference, affect the accuracy of theh-index.Aimed at the problem discussed before, a series of solutions to make optimization onh-index of authors in the case of huge amount of data is described in this paper. In order toachieve the sources of huge amount of data, making use of Web framework based dataextraction method to get huge amount data from Google Scholar and Microsoft AcademicSearch. Meanwhile a solution to avoid the author’s excessive self-reference is proposed.After that, the data we get before is integrated with DBLP and data from author’shomepage for later analysis use. Based on the data clustering graph theory, the relationshipof co-authors in the same articles as well as the publication information in authors’personal homepages are used to distinction different authors of the same name in thegreatest degree. All of the above provide a guarantee on the final computed h-index, whichcan get an h-index of more reference value for academic achievement evaluation.Based on the technology mentioned before, the academic evaluation system-AccEvais implemented in Jsp and Java. AccEva increases the amount of literature data from1.8million to9.5million, which is nearly4times larger than before. The functional testing ofthis system shows that AccEva is able to provide the basic information of the inquiredauthors including h-index, provide for users a guarantee on the precise orientation, andalso the proof of the author distinction. At the same time, authors’ literature lists andanalysis graphs of literature trend are presented on the pages of authors’detail information,which offer nice user experience. The performance testing of this system shows that theerror rate of authors’ literature list is less than5%and the h-index can reflect the author’sacademic achievement well. The average system response time is no more than1s.