Design And Implementation Of Paper Recommendation System Based On CiteULike

CiteULike(citeulike.org) is an virtual research publications sharing system. Unlike the other simple online library, the users are allowed to submit or post the publications which they like and meanwhile to add tags. Based on the users’above action, the system builds some relationships between users and publications. Accordingly, the users are recommended with related excellent publications by the system. On the event of accuracy and integrity, the system needs to be further improved currently. Compared with the other researching areas, such as recommendation systems for movies, books and music, the recommendation methodology of publications is still a virgin land. Therefore, it is becoming more and more necessary to build a content-based and collaborative filtering algorithm combined recommending technology, which extends the recommendation features.At first, the current features of the recommender system of CiteULike are discussed and its shortcomings such as cold start problem or the convergence of the recommended results are analyzed as well. And the necessity of improving the existing recommender system’s quality is also addressed. Based on these actions, the user requirement analysis for the recommender system of CiteULike is carried out and an idea is proposed for digging user’s characteristic out by suggested keywords and extending the recommendation result set by the combination of content-based filtering and collaborative filtering. Furthermore, the core processes of recommendation algorithm of CiteULike are designed and the framework of recommender system of CiteULike is also done. And the design and implementation of the algorithm is detailed as well. It is focused on how to implement the keywords recommendation with association rule mining and extending result set using clustering and genetic algorithms. At last, the advantages and disadvantages of the CiteULike recommender system are summarized.