Research On Personalized Recommendation Algorithm Based On User Interest Changes And Rating Differences

With the rapid popularity of the Internet in China and the rapid increase of Internet users,the amount of online information is exploding,and it is difficult for people to find what they want to obtain in the vast information.In order to solve such problems,personalized recommendation system was born.It collects users 'daily browsing behaviors and establishes relevant algorithm models to mine users' true preferences.The core content of the recommendation system is the recommendation algorithm.Collaborative filtering recommendation algorithm is one of the most widely used and most mature recommendation algorithms in the current recommendation system.However,as the amount of data increases and the scale of applications increases,the traditional recommendation algorithm also exposes many deficiencies.For example,the rapid expansion of data volume makes the data extremely sparse,which affects the accuracy of recommendations.Therefore,in view of these problems in the traditional collaborative filtering recommendation algorithm,this paper proposes a personalized recommendation algorithm based on changes in user interest and differences in ratings.The main research work of this article is as follows:(1)With the rapid expansion of the number of users and items,the rating matrix becomes extremely sparse,which affects the accuracy of the recommendation algorithm.To solve this problem,this paper improves the slope one algorithm to fill the matrix.First,perform k-means clustering on users.Then calculate the difference of item score in the target user cluster.Secondly,when scoring the target items,the influence difference and item similarity between items are fully considered.Finally,the matrix is filled.Experiments show that this improved algorithm effectively relieves the data sparsity and improves the filling quality.(2)In the past,recommendation algorithms often used only scoring information,and other useful information was discarded without making the recommendation results less accurate.In response to this problem,a collaborative filtering recommendation algorithm is proposed that merges user interest and rating differences.First,apply the TF-IDF idea to the user's weight calculation of tags.At the same time,in order to gain insight into the changes in user interest,it is captured by a combination of exponential decay function and time window.Secondly,according to the historical rating matrix,fully considering the impact factors such as the difference in user rating values,the difference in judging criteria,the difference in influence and the difference in project impact,a similarity measurement algorithm for rating differences is defined.Finally,the user's interest similarity and score difference similarity are weighted and fused to obtain more accurate user neighbors,which makes the target item's score prediction more accurate.Experiments show that the proposed algorithm improves the recommendation quality.(3)Combining the above two algorithms in sequence,a personalized recommendation algorithm based on user interest changes and score differences is proposed.In the designed collaborative filtering recommendation algorithm that merges user interest and rating differences,some of the data uses user ratings,and this part of the data has data sparsity problems.Therefore,before the calculation,this part of the data is expanded according to the improved slope one algorithm.Finally,the experiment through the dataset Movielens shows that the algorithm proposed before the data filling effectively improves the accuracy of the recommendation.