A Research On Trust And Distrust-based Collaborative Filtering Recommendation Model

Collaborative filtering is a widely used technique for the recommender systems, butit suffers from several issues such as low recommendation accuracy, data sparsity andvulnerable to attack. To address these issues, trust-aware recommender systems havebeen proposed recently. Distrust operater is introduced into the recommendation processto improvement the insufficient of filtering mechanism of trust-only recommendersystems. The main tasks are as follows with the background of movie recommendersystems.First of all, traditional collaborative filtering algorithms lack the time effect ofhistorical information. So we propose a similarity calculation method associated with adecay function of the rating’s produced time, and let the linear decay function as theweight to introduce the algorithm used to reflect the changes or drifts in the user’sinterest.Secondly, Calculation of the estimated value of trust and distrust is a key step in theprocess of collaborative filtering recommendation, the simple measure of trust anddistrust method does not consider the trust propagation and aggregation, we proposed ametric based on the weighted mean of trust and distrust. With the Storage method ofstack and queue, the trust value is calculated recursively.Thirdly, trust and distrust factor as a weighted factor or a filter factor. We combinedthese factors with the general method for collaborative filtering recommendation, anddivided them into the weighted mean and collaborative filtering types according to thecharacteristics of each algorithm. At the same time, in order to expand the coverage ofrecommended neighbors and improve recommendation accuracy, we propose a neighborcollection variable collaborative filtering algorithm and give the basic principles of it.Finally, the detailed experimental process is designed to verify the contents of thispaper. Through the contrast with the benchmark algorithm, we analyse the mean error,coverage and performance of our algorithm to prove that combine the temporal similarity,trust and distrust factor into process of recommendation can improve the accuracy and coverage.