Research On Recommendation Algorithm Based On Trust Enhancement And Dynamic Time Window

Thanks to the explosion of social networks,the recommendation system has become an effective way to deal with the problem of information overload.Traditional collaborative filtering recommendation algorithm utilizes user's historical data for recommendations,which is easy to face the problems of data sparsity and inaccurate recommendation.Therefore,many scholars try to integrate user's social relations into the recommendation algorithm.The research shows that the user's social information can alleviate the data sparsity problem effectively in traditional collaborative filtering algorithms.Meanwhile,most of them think that user's interest is immutable,and they don't take time as an important attribute to predict the interest of users.For deal with these problems,the paper proposes an recommendation algorithm based on users' social relations and the characteristic of interests changing with time.The main contents of this paper are as follows:(1)Because of the influence of social relationships on the recommendation results,this paper introduces trust relationships to build a trusted network and integrates users' attributes to enhance the trust relationships between users.Meanwhile,a recommendation algorithm based on trust enhancement and history behavior is proposed by using the user's historical rating information.Firstly,the breadth-first search algorithm is used to mine the indirect trust relationships between users,and then the dense trust network is constructed.This method can compensate for the sparse data.Secondly,implicit social relationships are introduced to make up for the deficiency of explicit trust relationships and further identify interested groups.Then,the explicit and implicit trust relationships are fused to obtain the comprehensive trust of users,and the user's attributes are combined to enhance the trust relationships between users to obtain the enhanced trust.At the same time,the final similarity is obtained based on the similarity of historical behavior and the degree of trust enhancement,which efficaciously boosts the precision of neighbors' selection.Finally,the items without a rating are predicted and the target users are recommended the items they are interested in.Comparison experiments on the Film Trust and Movielens show that the proposed algorithm is better than the other recommended algorithms in Recall,RMSE and MAE.(2)According to the characteristic of user's interest changing with time,this paper proposes a recommendation algorithm based on the Ebbinghaus forgetting curve and dynamic time window.Firstly,the algorithm simulates the change rule of the user's interest based on the forgetting curve and divides time windows by the time attribute corresponding to the user's rating.Secondly,considering the influence of the amount of data in time windows on the recommendation results,the algorithm integrates the concept of dynamic partition to ensure that the amount of data in each time window is greater than the threshold of number of information;and,short-term recommendations are made for the data within the time window.Finally,the time function is introduced to give the corresponding weight to the short-term preferences in different periods,and the final recommendation set is selected to push to users.Through the experimental comparison and analysis on Movielens and Netflix,the algorithm is better than the other recommended algorithms in Precision,Recall,MAE,RMSE,and F-Measure.