Research On Distributed Collaborative Filtering Recommendation Algorithm

In Distributed Recommender System, the user's profile information datas distribute on different nodes of the Network. Users cooperate with the others to generate his recommendation. And the recommendation generating does not need the information and computational recourse stored centrally. So it contributes to solve the problems that the central-server based recommender systems face: scalability, computational complexity, secret protection and so on.Distributed Collaborative Filtering Recommendation Algorithm for Distributed Recommender System is researched mostly in this paper.Firstly, aimed at the problem that the users are difficult to locate his nearest neighbors efficiently and the traditional formulas could not capture the similar relation of users enough, we proposed a Distributed Collaborative Filtering Recommendation Algorithm based the DHT protocol. In the algorithm, to improve the user's nearest neighbors searching efficiency, the users searched the information of his similar neighbors by the"fuzzy keyword"generated according the extreme rates in user's rate vector. And we introduce a weighting method to calculate the degree of similarity between users to improve the similarity calculation method according to the distributed situation of user's information data. When the weighting values are given, two factors, i.e. the propinquity degree between the ratings of users and the inverse preference frequency of the ratings themselves, are taken into account to reduce the active user's amount of calculation and increase the recommender precision.Secondly, aimed at the problems that the nearest neighbors of the users could not regroup efficiently when his information data changed in the Distributed Recommender System, we proposed a Nearest-neighbors updating algorithm in Distributed Recommender System. In this algorithm, to reduce the amount of calculation and communication in the process of the neighbors searching and to increase the user's nearest neighbors regrouping efficiency, we searched the neighbors for the user with his profile just changes and filter the neighbors with the interest similarity of users.Thirdly, analog experiments verify the effects of the Distributed Collaborative Filtering Recommendation Algorithm based the DHT protocol and the Nearest-neighbors updating algorithm in Distributed Recommender System respectively. The analog experiments verify the efficiency and precision of the recommender system according to the evaluation standards: amount of communication, quality of the nearest neighbors set and Mean Absolute Error and so on.