A Research Of Virtual Community Users Based On Collaborative Filtering Recommendation

With the exponential increase of the number of users and items,the amount of computation increases dramatically,and the collaborative filtering algorithm encounters performance bottlenecks.Therefore,reducing the collaborative filtering algorithm to search for nearest neighbor,improving efficiency and system scalability by clustering algorithm is the research hotspots.The traditional clustering algorithm inevitably faces the problem that the K value(clustering group)is difficult to determine and K value usually depends on people's experience.This paper finds that a co-clustering algorithm in the field of genetic engineering can effectively solve the empirical dependence of k-value.On one hand,Matrix after co-clustering is much smaller than the size of the original score matrix;on the other hand,users' scores within the same class have similar patterns and predicting is fast and flexible.Firstly,ISA(Iterative signature algorithm)is used to divide user-item matrix(rating matrix)into the overlapping groups instead of restricting one user just belong to one group,so we can describe user preferences more comprehensively.Statistical analyses show that,clustering is better.Secondly,the virtual group user that fuseing the interest of each group based on virtual user technology is constructed as the input basis of collaborative filtering.Considering the fact that different users have different investment time for different projects,this paper constructs a virtual group user based on expertise degree weighting.The effectiveness of the strategy based on professional degree is verified compared with the mean strategy and the median strategy.Finally,the algorithm effect is verify on the Filmtrust data set.The experiment shows that the algorithm can improve effectively the accuracy of collaborative filtering algorithm.