Collaborative Filtering Recommendation Algorithm Based On Dynamic Trust Model

With the rapid development of the Internet,users and network size are showing exponential growth.Large numbers of information resources result in users' obtaining data is faced with enormous challenges.Even though search engine technology allows users to find specific information,it's only available to users with the same search results,while cannot meet the user's personalized service requirements.Personalized recommendation systems alleviate the problem of trust overload effectively.Recommended systems not only helped users to obtain personalized information,but also helped some businesses to promote their products,which brought significant economic effects.Among all type of recommended systems,the collaborative filtering recommendation system is the one that was used most widely and which proved to be the most effective.But it also has its own shortcomings which affect the accuracy of the recommendation systems,such as data sparseness problem,cold start problem,the problem of poor expansibility as well as the defenselessness for malicious attack problems and so on.Aiming to solve the issues exist in the collaborative filtering recommendation systems,this paper recommended the user's trust degree with time sequence into the matrix factorization recommendation model,according to which an improved algo-rithm based on dynamic trust model--Sequential Trust MF(STFM)recommendation algorithm.In this paper,the main work is as follows:(1)I design a dynamic trust model based on the cognitive.In this model,trust is generalized as total trust,direct trust and indirect trust.Through the historical evidence window,the total trust degree is changed into direct trust or indirect trust.In the calculation of trust degree,the time series information is introduced in the calculation of the direct trust degree,which makes the trust model dynamically compute the trust degree between users.The calculation of indirect trust is based on the Trust Tree Direct(DTT),and it is realized by polymerization method which is constructing the indirect trust propagation degree.With the confidence of the strength of the distinction,trust is no longer a simple two element trust in this model.And the final trust degree can be adapted to the change of the relationship between the network and the network to a certain degree.(2)To calculate trust by the trust model,and to introduce the user's trust degree with time sequence into the matrix factorization recommendation model,according to which an improved collaborative filtering algorithm based on dynamic trust model--Sequential Trust MF(STFM)recommendation algorithm--is proposed.And carry out experiments on Epinions dataset.we finally prove that this algorithm can improve the recommendation accuracy obviously compared with other recommenda-tion algorithms separately at the point of trust users number,parameters factor and the number of product recommendations.