Research On Bipartite Network Structure-Based Recommendation Algorithm

With the continue development of network technology and e-commerce, information also presented surplus state, which makes the users can not find the information which satisfy their needs from the expansion of information in line. In this background the recommendation system is proposed, recommendation algorithm is the core of recommendation system, and bipartite network structure recommendation algorithm attract more attention because of its low recommendation complexity, the recommended content is not restricted, and its diversity.By analyzing the bipartite network structure recommendation algorithm advantages and weaknesses, the main work of the thesis is as follows:Firstly, for the problem of single issue recommended items of network structure recommendation algorithm, as it neither determine whether the user selected over projects nor care the user preferences influences on resource allocation, and it depends more on the size of the project and users’ node, resulting in a single recommended project, one combining user preferences and similarity of network structure recommendation algorithm is proposed. On the basis of the bipartite network structure recommendation algorithm, introducing user’s preference scores, using maximum and minimum methods to standardized user’s different ratings, as the weights of network architecture. In the first resource allocation process, introducing the ratio of the degree of project and the weight of project,enhanced the recommended capacity of the minimum weight of project. In the second resource allocation process, introducing user similarity, increasing the proportion of the resource allocation. Through the twice of resource allocation, more in line with user’s preferences of project were incorporated into the front of the recommended list.Secondly, bipartite network structure recommended in the presence of a user cold start problem, on the basis of the trust relationship between users, we propose a recommendation algorithm fusion trusted network and bipartite network. The user similarities are calculated according to the trust ratings between users, and the set of users with high similarity is created for each user, and the degree of trust between different users is differentiated, then the new trust relationships are predicted by using the differentiated degree of trust and the sets of similar users. The user similarities are calculated according to the bipartite network structure recommendation algorithm, the user’s similarities as the implicit trust value, through normalized the two trust values,easing the new users’ cold start problems.Finally, the two improved recommended algorithms are made detailed simulation experiments, experiments show that two improved the recommended algorithms has better accuracy and diversity, the research on this thesis has higher value.