Research On Collaborative Filtering Approach Based On Initial Trust

With the rapid development of computer network technology, information overloadproblems have become more serious. As an effective method to solve those problems, therecommender technique plays an important role. Collaborative filtering recommendationsbased on user rating data have been the most widely used today. However, as the numberof users increasing in the recommender system, the strange users who rate rarely appear.Meanwhile, the traditional collaborative filtering technology also ignores the trustrelationships which are important factors for recommendation. Above two problems haveled to lower recommendation quality. In response to those problems, on the basis ofdomestic and international research, combining with the initial trust between users, thisarticle on how to further improve the quality of collaborative filtering recommendation isstudied.Firstly, the existing trusted methods required direct historical information about thetarget user to calculate the value of their trust, and can not calculate the trust value of theunfamiliar users who are lack of historical information or none. Combined with theestablishment of trust between users, we propose a initial trust model. Filter user historicalinformation to build the target user interest grouping similar user rating matrix. Use Bayesmethod to calculate the direct trust between each group and user, and the initial goal of thetrust is gotten by using the weighted average method based on the time and number ofinteractions. Build the calculation method of the initial trust based on above process.Secondly, regarding on existing collaborative filtering algorithm recommendationlow accuracy and coverage of small issues, we join some strange users who don’t have theindirect trust relationship with others into the original trust network based on initial trustmodel. More indirect trust between users is obtained by the proposed trust deliverystrategy in the text, which matches more neighbors for users. Generate a matrix of trustusing the known trust information, which reduces the sparse degree of data，and ultimatelycomplete the recommended target users by the user similarity matrix combined matrix andtrust. Following the above process, we propose the collaboration filtering recommendation algorithm based on initial trust network.Finally, we conduct experiments on the comparison and analysis of the proposedmethods and other related methods by using the data in Extend Epinions dataset.