The Application And Research Of Topic Model On Paper Recommendation

With the rapid development of technology, scientific papers are increasing with rapidspeed, which provides abundant literature to researchers. According to related statistics,university papers growing at an annual rate of6%to8%has reached scale of millions,making the searching time growing rapidly.Based on this background, paper recommendation system emerges. Traditionalrecommendation system based on keyword matching or paper meta-data can’t get semanticsof paper and inefficient, recommendation result is not ideal. In order to solve the problem, thispaper presents a new recommendation method based on topic model and proposes a newalgorithm for user interest model. In this paper, the main work is as follows:1. We analysis the related theory and research situation of topic model and paperrecommendation system, and analysis the feasibility of topic model in the field of paperrecommendation system.2. We put forward a new user interest model TV-IPF, the algorithm increases the valueof less frequently, high weight of user interested paper list in the user interest model, at themeanwhile not lowering the theme appearing in most of user’s interesting list. Compared withthe traditional method of mean algorithm, based on the improved algorithm of user interestmodel and recommendation results fit more with user’s real demand.3. We design recommendation system based on topic model, examine the effects of thetopic model in solving the problem of semantic, analysis the influence of sparse andrecommendation number is studied. In addition, this paper analyzes advantages anddisadvantages of collaborative filtering algorithm and topic model, then a mixture algorithmsystem put forward, based on the sparse degree to determine the proportion of these twoalgorithms in the recommended list. The mixture recommendation system can provideaccurate recommendation in sparse matrix, meanwhile increasing the diversity of system.Experiments show that the model improves recall and diversity of recommendationsystem.