Personal Recommendation Based On Collaborative Filtering

With the rapid development of internet, we come into an information explosion society. Faced with massive amounts of information, we find it difficult to seek what we really want. Recommendation system builds the binary relation between users and objects and makes use of the existing selection process to dig everyone’s potential interested objects.A lot of recommendation algorithms have been proposed. Collaborative filtering is firstly proposed and widely used. But there exist some defects in the collaborative filtering algorithm such as rating-sparsity, bad-extendibility and precision-deficiency.To solve the problems, the users or items can be regarded as dependent nodes that are extracted from their contents. Then we explore the community partition on the user-item bipartite network to utilize the similar users or items. Based on the community partition, we can make the personal recommendation for target users. Therefore, this paper comes up with a new collaborative filtering way which is based on the community detection. This contains three stages: data formulation, similarity calculation and recommendation. In the first stage, the apparent relationship between users and items are formulated with bipartite network instead of traditional user-item rating matrix. In the second stage, the bipartite relationship formulation is regarded as a topological network. We can detect community splition through bipartite modularity maximization. Similar users or items are splited into the same community. In the recommendation stage, we offer two recommendation mix lists based on the user-community recommendation list and item-community recommendation list. According to the scores, we make top-N items as the final recommendation items for target user.To solve the extendibility problem, this paper realizes the collaborative filtering recommendation algorithm based on item similarity and community detection using Mapreduce programming structure which is built on the hadoop platform. This algorithm consists of three procedures: data pretreatment, recommendation algorithm realization and performance evaluation. The precision, recall rate and F measure and used to give a comprehensive evaluation to algorithms. The execution efficiency can be highly improved and the time complexity can be reduced through the massive data processing way. The application of Mapreduce can also extend application bound.