Personalized Recommendation Algorithm Research Based On Jaccard Item Category Similarity

With the ever growing importance of internet, people are becoming overwhelmed by information. Recommender system is designed to help people to find the most relevant items based on their preference. The most widely used technique is collaborative filtering, which recommends products to users based on the experience of others. But CF suffers from profound problems such as the sparsity problem and scalability issues. This thesis studies how can overcome the problems, while achieving better prediction accuracy.Firstly, the research background of the thesis and some personalized recommendation techniques are briefly introduced, and collaborative filtering is emphatically introduced, the problems which faces are analyzed.Secondly, aimed at the sparsity of user rating data and the drawbacks of Traditional similarity measure methods, a method of computing item rating similarity based on cloud model is proposed. The method compares the similarity on knowledge level, which can overcome the drawback of attributes'strictly matching of traditional vector-based similarity comparison method. The method solves the problem the traditional collaborative filtering algorithm facing, and improves the effectiveness of the recommendation.Finally, aiming at the problem of considering factor singleness when computing item similarity of item-based collaborative filtering, a computing item category similarity method based on Jaccard similarity is proposed. It first analyses the personalized recommendation algorithms which have not considered the category factor and points out the shortcomings. And then it introduces some methods having considered the factor, and analyses the deficiency of the methods computing item category similarity. The method of computing item category similarity based on Jaccard similarity is proposed at last. Combining item category similarity and rating similarity can get the combined similarity. It chooses the top-k items of the highest similarity as the nearest neighbor of the target user, and then predicts the vacant items'ratings of the target user and implements the top-N recommendation. The experimental results show that the algorithm can not only efficiently lower sparsity of rating data but also improve the recommend quality of the recommender system.