Personalized Query-Oriented Reference Paper Recommender System

With the spring of the huge number of publications in real world, it is animportant and challenging issue that how to alleviate the overload of the academicinformation and provide a researcher with an efficient and accurate recommendationof reference papers about his/her interested field or topic. It is significant andchallenging to recommend the reference papers to a research in the massive amountsof information. In this article, we address the personalized query-oriented referencepaper recommendation. For this task, we proposed two important measures, that is,query relevance and time dependent paper impact. The former looks from theresearcher’s point of view to measure the relevance between query and papers, and aunified graph model is proposed to handle this, in which it integrates various types ofinformation, including papers, their content, topics, authors, as well as therelationships among them. The latter mainly focuses on the quality of the papers byexploring their impact influenced by time in the citation graph and is used to filter outthe ones which are important for researchers.These two measures are both important for recommending reference papers. Weconduct a series of experiments to evaluate the proposed methods on the ACLAnthology Network (AAN) datasets, and significant results are achieved. In addition,in our algorithm, the experimental results show that the query relevance is moreuseful than paper impact function, and personalized query relevance is clearly muchbetter than the non-personalized query relevance. On the other hand, thetime-dependent paper impact is more helpful than the time-independent paper impact.