The Research And Improvement Of Collaborative Filtering Recommendation Algorithms

Information overload is one of the most critical problems, and personalized recommendation system is a powerful tool to solve this problem. However,there are so many kinds of shortages in traditional recommendation methods that the recommendation system cannot reach a good result. So the paper firstly analyzed and studied the existing methods and based on its existing problems, then puts forward an improving method. The major work is as following.First of all, familiar with the relevant skills and knowledge of recommendation system. Similarity functions in traditional CF compute a similarity between a target user and the other user without considering a target item. The similarity between a target item and each of the co-rated items should be considered when finding neighbors of a target user. Additionally, a different set of neighbors should be selected for each different target item. In the new similarity function, the rating of a user on an item is weighted by the item similarity between the item and the target item.Secondly, since the cold start problem that troubled recommendation system for a long time cannot be solved effectively, the paper proposed a new algorithm to solve this problem. when the number of rating records is small, but not zero, by developing a new similarity measure for collaborative filtering systems. One advantage of this approach is that no additional information is required other than the rating data that collaborative filtering systems basically use and that existing collaborative filtering recommendation systems can be easily updated by only replacing the similarity calculation part.Finally, investigated the similarity of different items and cold start problem for experiments, through collecting user rating of some of the items, choose appropriate experiment validation data sets. And with some existing algorithms in the experiment analysis and compare the results, the experimental results show that the proposed method can reflect the thinking of the new algorithm and the running result is reasonable, effective and feasible.