Research On Personalized Recommendation Method Based On Context And Social Network

With the advent of the Web2.0 era, the biggest problem we are facing is the information overload, search engine and recommender system appears to solve the problem. Collaborative filtering is one of the most commonly used as well as the most successful recommendation technology, and it has been widely used in many fields such as e-commerce. But the traditional collaborative filtering recommendation algorithm has some problems such as sparsity and cold start, therefore, it is very necessary to introduce a new data source in the recommendation algorithm to alleviate these problems. Context and social networks can provide more information about the user to the recommender system. Compared to the traditional collaborative filtering recommendation algorithm, the introduction of context can help users more accurately find the project they are interested in, and the social networks can provide more user preference information and user behavior data, to partially alleviate the sparsity and cold start problems of the recommendation system. So how to fuse context and social network information to improve the accuracy of the recommendation system is a problem worthy of study.In this paper, several recommendation algorithms which fusing time context, location context and user social network information are proposed. Based on the similarity calculation of traditional collaborative filtering algorithm, we make the following improvements:1. The user’s interest will change over time, we consider two aspects of time: one is the user interest similarity in time variation, another is the similar user’s recent interest. So a time decay function is introduced to reflect the change of user interest in different periods.2. There are some differences in the interests of users in different places, so we classify users according to different cities, users in the same city location similarity weight is significant, the different city user’s location similarity weight is small.3. Users tend to be more trust the recommendation information based on their own friends, so if the recommended item is the item their friend interested in, the user will be more trust the recommended results. We calculate the user’s interest similarity according to the user’s social network information, the higher the proportion of common friends between users, the more familiar among users, the higher the similarity of their interests.In this paper, we put these three kinds of information into the traditional collaborative filtering algorithm by linear weighted fusion in different ways. Experiments on the Douban dataset show that the proposed algorithms have a certain degree of improvement in the accuracy compared with the traditional collaborative filtering algorithm.