The Reaserch Of Recommend Algorithm And Strategy In Recommendation System

With the development of the web2.0technology, the Internet applications become more and more popular in our daily life. At the same time, the exponential increase of the resource makes the information overload become a serious problem. To help users in making decision, recommender system has been designed.Different from the traditional search technologies,recommender system is based on the users’hi story record to generate intelligent recommendations, therefore, recommend technique quickly became a hot topic and is widely used in social networking and business website.Collaborative filtering (CF) is the most important recommend technique. It is based on the similarity of users’ behavior to recommend same items. Generally, there are there challenges in the research field. The first is to solve the sparsity of user data. In fact most users do not purchase or rate most items and thus there is a huge sparsity relationship between users and items, which affects the accurate of the result. The second challenge is to solve the cold-start problem.It’s difficult to get the relevant information as the new users and new products added to the system. The last is to improve the accuracy of recommendations. How to dealing with the sparsity of user-item matrix and build the users’model, is still a hot topic in the current study.In response to the challenges, this paper makes the main contributions and summarized as follows:Firstly, we make an improvement of collaborate filtering by the time weight iteration model. In fact, users’interests always change with time.However, traditional CF algorithm does not take the time factor into account and hence cannot reflect the changes. In this paper, the change of users’interests is considered as the memory process, and a time weight iteration model is designed based on memory principle. For a certain user, the proposed model introduces the time weight for each item, and updates the weight by computing the similarity with the items chosen in a recent period. In the recommend process, the weight will be applied to the prediction algorithm. Experiments show that the new algorithm can improve the accuracy of the recommend.Secondly, we propose a new recommend strategy combined with rough set and fuzzy theory. Rough set theory is responsible for the reduction of attributes sets and getting the preference rules of each user, and output the result in the form of a matrix in the recommend system. While the membership of items to the attribute sets is calculated by fuzzy statistical method, and also export as a membership matrix that kept in the database. Recommend system will make the synthesis operation on the rule matrix and membership matrix, to get the relationship matrix of users and items, and then, system will calculate the recommendations based on this matrix. Finally, the experiment on the MovieLen’s data set proved the effectiveness and veracity of the algorithm.