A novel approach for paper recommendation based on rough-fuzzy set theory

Abstract

Nowadays, recommending relevant and valuable academic papers has drawn great attentions from academic researchers. However, most of time paper recommendation still relies on the keywords given by users as input. Previous studies have shown that this kind of method are not effective because they did not take semantic similarity into consideration. To address the problem, in this paper we conduct similarity analysis based on synoptic content including the titles and abstracts between the papers referred by users and the ones in paper dataset, and then recommend scientific papers to the users. We use the TF-IDF to pick out the important words of the titles and abstracts in papers, and apply rough-fuzzy set method as well as WordNet to calculate the similarity values of candidate papers. After ranking the papers based on the values, we can get the paper recommendation result. The experiments on the public available dataset have demonstrated the superiority of our proposed method over other baselines.