Research On Collaborative Filtering Recommendation Algorithm Combining Time Factor And Contribution

With the explosive growth of information today,recommendation systems can provide users with personalized recommendation services in the vast amount of information.It generates recommendations for target users by using different recommendation algorithms.Among the many recommendation algorithms,collaborative filtering algorithms are most widely used.However,with the increase of users and items,the algorithm faces some problems.This thesis focuses on the problem of that traditional collaborative filtering algorithm does not take into account user interest changes,excessive differences in ratings of the same item by different users,the influence of popular items and asymmetric influence among users to research solutions and design more efficient recommendation algorithms.The specific work is as follows:To the problem that the traditional user based collaborative filtering algorithm does not consider the change of users’ interest and the excessive difference of different users’ ratings for the same item,a collaborative filtering recommendation algorithm combining contribution and time factor(rating time)is designed,named CTCF,which introduces the credible error threshold,contribution and time weight in the calculation of user similarity.First,the user-rating matrix and user-rating time matrix are constructed using the user rating information,and the user contribution is calculated based on the credible error threshold;then,a forgetting curve is introduced to fit the contribution and rating time to obtain the time weight,and then the time weight is introduced into the Pearson correlation coefficient to calculate the user similarity;finally,the nearest neighbor set of the target user is found,and the items that are liked by the nearest neighbor but not rated by the target user are filtered,and the target user’s ratings for these items are predicted,and Top-N recommendations are generated from the high to the low rating.The test results on the Movie Lens dataset show that the CTCF algorithm has higher F1 values,which effectively improves the recommendation accuracy and dynamics.To the problem that the traditional user based collaborative filtering algorithm does not consider the influence of popular items and the asymmetric influence among users,a recommendation algorithm is designed to combine influence and CTCF,named ICTCF,which is based on CTCF and reduces the influence of popular items by introducing a penalty function based on the number of item interactions;uses indicators such as credibility,reliability,and self-awareness to characterize user characteristics,and combines them with the penalty function to calculate user influence;integrates user influence and user contribution to obtain a new contribution,and improves time weights with the new contribution;uses it for calculating user similarity.The experimental results on both hetrec2011 and ml-latest-small datasets show that the ICTCF algorithm can improve the recommendation quality.A word memorization recommendation prototype system is developed,and the designed ICTCF algorithm is used in the system to generate recommendation lists.The application results show that the research results of this thesis have certain practicality.