Research on Citation Recommendation Strategy Based on Multiple Factors
|Course||Computer Software and Theory|
|Keywords||citation networks clustering result recommended|
With the quickly promoting speed of information spreading, the number of available scientific literature is also rapidly increasing. It is a very difficult task for users to find the results they need among thousands of citations. Recommendation is one way to solve this problem. It can predict the items which users may be interested in and recommend them to users. For the reason that the similarity of the citation titles can not accurately reflect the similarity of the citations, the existing result recommending methods cannot give satisfying recommendations of citations.According to the characteristics of citation, we propose a new citation recommending strategy based on multiple factors, which includes self-citation factors and user factors. First, we can generate a citation reference graph based on the reference relationships of citations. Then, we define a series of rules according to the relations between citations. In view of these rules we assign weights to reference edges to express association strength. Then we cluster closely linked citations with a given clustering algorithm. According to the clustering results, we identify relevant citations users need, and generate the initial citation recomending set. Based on the behavior of users which are similar to the current users’, we adjust the set generated before, and finally generate the final recommending citation set.The thesis mainly studies the following issues. a) Combine the item-based strategy with the user collaborative filtering strategy, and propose a citation recommending model. At the beginning when users’ratings are sparse, we recommend using the similarity of items. After the system runs for a while, adjust the similarity based on users’ evaluations. b) Instead of the traditional semantic similarity, the strategy calculates the similarity of items according to the properties of citations, so that it can avoid extraction and semantic analysis errors. c) The thesis improves the generation method of user-item rating matrix and solves issues in traditional user collaborative filtering strategy.In this thesis, by comparing a large number of experiments, it is verified that the proposed citation recommendation method based on multi-factor performance is superior to the existing recommendation methods. So in the WebCitation system, the produced sets of citation recommendation on the target citation can meet the needs of users and reduce the burden on the user’s operation.