Research On Collaborative Filtering Algorithm Based On Weighting Matrix Completion

With the booming development of the internet technology and the information technology, the personalized recommendation technology is going gangbusters. The Collaborative Filtering recommendation algorithm, as one of the most famous personalized recommendation system, is also one of the most studied algorithm, the current and the future a period of time, will become important research aspects of e-commerce filed. Along with the expansion of application scope and application of changes in the environment, combined with the actual site of the requirements on the accuracy and efficiency of recommendation. Some disadvantages of the collaborative filtering algorithm are appear, which typically have a recommendation real-time problem, the new user or new item of cold start, the accuracy of the recommendation results, etc.When calculating the similarity between two users or two items, the traditional collaborative filtering algorithm all just by user rating information for the item. Rarely take into account the effect of other information. And using the existing information as much as possible to predict the result for recommendation system, plays a crucial role to improve the accuracy of the recommendation algorithm. In this paper, by utilizing the users’ rating time information and the item itself category tags, we improve the collaborative filtering algorithm accuracy and recommendation efficiency, and researches are done as follows:Firstly, through the analysis considering the influence of time factor on the collaborative filtering algorithm, we put the user-item rating time into the user-item rating value, that’ s to say, make the user-item rating value of the original data set a weight correction with the rating time information. Then propose a collaborative filtering algorithm which fusion time delay curve.Secondly, in view of the traditional Item-based collaborative filtering algorithm in calculating the similarity between items, without considering the items’ own attributes influences on items’ similarity. This paper proposes fusion the degree of difference between items into the formula for calculating the items’ similarity. By predicting the target user unrating item, complete the user-item rating matrix. then according to the size of prediction rate, give appropriate recommendations for the target user.Finally, we apply the improved algorithm to dataset Movie Lens, design experiments. Finding out the smaller MAE between the cosine similarity and the pearson correlation coefficient when calculating the similarity of items. Then by contrast the traditional algorithm, it is proved that our proposed weight adjustment for matrix completion when calculating the similarity between items improve the recommendation accuracy and efficiency.