Research On Recommended System Of Scholar Paper Based On Topic Model

Academic paper recommended is an application of recommended system, that can help user find the paper they want from millions of papers. It can mine user’s latent re-quirements by analyzing user’s accessing history and recommend similar content to make search paper easily and expedite.Considering the specificity of recommended item, evaluating recommendation result with accuracy metrics merely can’t content different requirements of users. This experi-ment set improving serendipity of recommendation result as main objective. Serendipity represents the ability of discovering the items which is not known but needed by users.This paper employs the method based on topic model, with the words distribution of document is known, compute the posterior probability of topics and mine its latent struc-ture. It is not like the traditional space vector model, considering dimension on dictionary space merely. This simple method based on word frequency statistics can’t catch statis-tical characteristic and semantic characteristic. Topic model introduce the topic space, achieve the signify of document on topic space. Not only catch the semantic information to mine the latent relation, but have remarkable interpretation for recommendation result.The main content of this paper has three parts. LDA topic model based method is recommending based on distribution statistic on topic of document. The method which is at topic level can effectively solve the problem of polysemy and synonymous which is important to analysis document content and extract eigenvector. And comparing with TF-IDF based method, LDA topic model except achieving more surprised recommendation result. CTM correlate topic model based method considering the relation between topics and can effectively mine the latent relationship of document to improve the serendipity of recommendation. Relevance feedback based method adds the user rating, system can form the user profile, thus give different user the different recommendation, make the recommendation more personal.