Paper Classification by Topic Grouping in Citation Networks

The enormous popularity of Web 2.0 social network services has led to much research on social network analysis (SNA). These studies focus on analyzing the complex interactive activities between users in the world of virtual networks. SNA has shown great potential in automatic document classification, especially in identifying citation networks of research papers and the references among them. This research adopts the Clique Percolation Method (CPM) to identify all overlapping subgroups in a citation network. In the grouping process, research papers with similar topics will be grouped into the same topic group. Two papers are regarded as having a relationship when the common citation rate between them is higher than the threshold. A modified TF-IDF calculates the weight of each keyword in the topic groups. The keyword-weight vector represents the main features of each group, while the category of a new-coming document is determined by a novel similarity function. All the papers under study are collected from the journal IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) published from 1979 to 2011.