Research On Group Recommendation Algorithm Based On Probability Model And Feedback Mechanism

Recommendation system,one of ways to solve information overload,to some extend ease the dilemma between convenient life and complicated information with the innovation and development of internet technology.Now-days these systems have been applied in lots of known web sites,such as Amazon,Taobao,iQIYI,Sina microblogging and etc.,and having achieved satisfactory feedback.However,it is not considering user's demand variation in the systems that an obtained result merely concentrates on rather a single user than multiples ones,let's say family outing and friends gathering.Therefore,to recommend for the group users becomes a hot spot in web recommendation area.Due to neglecting the fact that the user's preference in group possibly takes a change,the performance of traditional recommendation algorithms suffers.To solve the problem,given a above scenario,the group recommendation algorithms are proposed from two aspects of feature matching and feedback mechanism to achieve a mere accurate and efficient result.On the one side,we have studies,on the feature matching and characterization of users and items properties respectively.A personalized recommendation list is obtained for group as follows:(?)mach the relevance according to related concept about feature matching,due to item-user attractiveness and user-item rating;(?)a matching probability model is established by the method of feature matching;(?)generate a personalized list by means of calculating the correlation probabilities between users and items and them sorting them following the ranking strategy of personal recommendation;(?)a personalized recommendation is extended to a group recommendation between of the fact that there is a correlation between the group and the item i if a member of the group is related to the item i.On the other side,we focus on feedback mechanism for group recommendation algorithms.The key problem is that how to dynamically adjust the preferences of the group users aims to meet the overall maximization satisfaction of group in the process of group recommendation.More specifically:We introduce a four large architecture model auel distribute weight in accordance with users'roles.A feedback model is presented to adjust the preferences of group members,whose recommended strategies weight-based aggregate voting and weight-based least misery andwhose evaluation criteria are the overlap similarity and Hamming distance.In the end,the two algorithms,namely a group result generation algorithm and a feedback vector generation algorithm,are proposed.In addition,the results obtained by an iterative computation meet the requirement of maximizing satisfaction of group members.The algorithms developed in the paper have been implemented in the practical projects.All experiments with Movielens as data-sets ralidated effectiveness of the algorithms based on probabilistic model and based on feedback mechanism respectively.