Research And Optimization On Collaborative Filtering Recommendation Algorithms For Data Sparsity

With the rapid development of Internet and the extensive application of e-commerce, the network was flooded with more and more information and service. While the users can enjoy the convenient and fast information service that e-commerce brought, they can also lost in the information space, thus resulting the problem of "information overloading". Therefore, personalized recommender system in E-Commerce came into being.Personalized recommender system in E-Commerce is based on users’ personal habits and preferences to recommend the program of information, goods, or services to users. In e-commerce platform, the personalized recommendation system play the role of the sales staff to provide users and recommend the products that interest to the user, helping users successfully accomplish their shopping.Recommendation algorithms is the most critical technologies in the recommendation system. Currently, collaborative filtering recommendation algorithms is one of the most widely used and most successful technology in recommendation system, it mainly consist of recommendation algorithms which is based on the user and the recommendation system. However, with the expanding scale of e-commerce, collaborative filtering recommendation algorithms encountered some challenges, such as data sparsity, scalability problems, the cold start problem and the time factor. This essay try to research in a high level according to the traditional collaborative filtering recommendation algorithms, improving its way of thinking and obtained some results.The main study of this thesis as follows:(1) It has a deep study about the e-commerce personalized recommendation system, including e-commerce personalized recommendation system development process, architecture, and recommend the technology used in the system in detail, the final e-commerce personalization recommendation system, the role of it.(2) It studies and analyzes the traditional collaborative filtering recommendation algorithms, including user-based and item-based collaborative filtering recommendation algorithms. Focusing on the implementation steps of the algorithm, pointing out the shortcomings of traditional collaborative filtering algorithms, but also introduce some improvements.(3) Proposing a collaborative filtering recommendation algorithms improvement, and this is a core part of this article. Based on the user-based collaborative filtering recommendation algorithms to recommend the idea of the algorithm for data sparsity, based on the user, through a item similar sets to calculate the prediction score of the item did not score, fill the user-item rating matrix, there by reducing the data sparsity. The introduction of a function of time for the time factor, calculate the prediction score, score time score gives greater weight away to score time score to give smaller weights to improve forecast accuracy.(4) Simulation tests. Matlab software to achieve the improved algorithm, and use Movielens data set to test the algorithm, verify its rationality. This article first compares the similarity of the three similarity calculation method of cosine similarity, correlation similarity and adjusted cosine similarity, and then this into the collaborative filtering recommendation algorithms and the traditional collaborative filtering recommendation algorithms were compared with experiments. Experiments to improve the algorithm achieved a certain effect, to improve the quality of recommendation.