Community discovery for academic papers using both citation and attribute information

Abstract

The main goal of this thesis is to discover the high-quality communities for academic papers. Sincetremendous amounts of academic papers have been published, many researchers experience diculties inexactly finding the papers in which they are interested. Community detection (or cluster discovery) canfacilitate this task since it finds similar or relevant papers. Thus, community detection from academicpapers has received a lot of attention, but it is still a challenging issue. Along this direction, citationanalysis and attribute (e.g., content) analysis have been widely used. However, most existing methodsfocus on either citation analysis or attribute analysis, disregarding the other side. The novelty of thisthesis is a complete merger between citation analysis and attribute analysis. Our approach constructsa network of academic papers by considering both types of information together and then performsclustering to obtain the communities of papers. In the network, an edge between two papers is createdby considering both (i) the existence and importance of citations from one to the other and (ii) theattribute similarity between the two papers. In this way, the two types of information are considered atthe same time, not sequentially. The optimal merger was empirically determined. Last, the effectivenessof our approach was verified by extensive experiments. About half-million papers were crawled, and thefull text was extracted from them for attribute analysis. The results show that our approach produceshigher-quality communities compared with the baseline approaches that use either citation analysisor attribute analysis. Overall, we believe that our approach will be very useful for academic searchengines.