The Research Of Literature Recommendation Based On Multiple Interests

With the explosive growth of the total amount of scholarly papers,it is increasingly difficult to find literatures interested by a researcher in such a large library.Scholarly paper recommendation is an effective way to overcome this issue.The present researches on the recommendation of scholarly papers are mainly concentrating in content based filtering,citation network,co-authorship network and paper evaluation index.The content based filtering method establish users' model by their historical operations,comments,interest labels and other information.However,this recommendation method requires a lot of time to collect above mentioned information.The recommendation method based on citation network utilizes the reference relationship between papers to recommend papers to users.But,the uncertainty of the citing and cited relationship between papers sometimes affects the quality of the recommendation results.Co-authorship network based recommendation is a kind of method which utilizes the complex network formed by scholars to recommend.The method based on paper evaluation index filters and recommends papers to users by citation and co-citation quantity,quality factors of journals and H-index of papers or scholars.On the basis of the existing researches,this thesis proposes a scholarly paper recommendation algorithm based on multiple interests.Our main contributions are as follows:(1)Identifying the multiple research interests of a scholar.We use clustering algorithms to divide the published papers of a scholar into several interest sets according to the fact that scholars usually have many research directions.(2)We propose two kinds of multiple interest scholar models,one model is based on VSM and another is based on frequent patterns.The VSM based model combines all published paper models in a same interest set by weighting,and take the combined feature vector as the corresponding interest model;the model based on frequent patterns preprocesses an interest set by LDA at the first,then mine a frequent pattern set from the preprocessed results by FP-Growth algorithm,finally simplify the frequent pattern set to build the corresponding interest model.(3)We also proposes the concept of importance degree of an interest,and gives the calculation formula of importance degree through the number of the papers in an interest set,at the same time,we introduced importance degree into the two proposed multiple interest scholar models.Three group experiments are conducted in a real dataset.The experiments results show that compared with the existing scholarly paper recommendation algorithms,the proposed algorithm can improve recommendation accuracy.