Academic User Interest Extraction using Multi-feature TextRank Based on Interest Attenuation

Abstract

User interest extraction is a bask task in the academic user behavior mining field. This paper proposes a multi-feature TextRank algorithm which combining behavioral characteristics of academic user and word characteristics of academic resource, aiming to extract academic user interests. This paper selects the real user behavior data from Agricultural Sci&Tech Information Resource Co-construction and Sharing Platform to conduct the experiment. Compared with TextRank, multi-feature TextRank improves P value by 4.8%, R value by 3.3%, and F value by 3.7%. And each performance of multi-feature TextRank is better than TF-IDF algorithm. The experiment shows that the multi-feature TextRank improves the accuracy and recall rate of TextRank, and it can be applied to automatic extraction of user interests in the context of multi-source academic resource usage.