Collaborative Filtering Algorithm Based On User Preference And User Clustering

At present,the strategy to solve the problem of "information overloading" on the Internet is mainly a personalized recommendation system,in which the collaborative filtering algorithm is one of the most widely used technologies for recommending systems.Collaborative filtering algorithm provides users with many conveniences in practical application,but also faces the challenge of how to mine the user's accurate and real interest preferences in sparse data and how to reflect the user's interest migration in real time.These questions have a greater impact on the quality and credibility of the recommendations.Therefore,this thesis studies and practices the above problems in detail.The specific research works are as follows:For the problem of data sparsity,this paper proposes an algorithm based on the user's real preference.The algorithm focuses more on the items given by users with high scores when they are interested in the real users.Based on the number of user's visit,the algorithm introduces the feature matrix of user's item scores based on user ratings.According to this feature,the score of each feature of the rating matrix is greater than the average number of users to analyze the score from the subjective ratings of users to accurately draw the user's real preferences.For the current user interest migration problem,this paper further proposes an improved algorithm based on Logistic function.The algorithm mainly deals with the problem that the user's interest migrates over time.The algorithm mainly monitors the change of the user's interest based on the time factor in real time.When analyzing the user's real preference,the algorithm uses Logistic function to introduce the time-weighted calculation to distinguish the user's interest in different time periods,increase the weight of the information of the recent user interest,and weaken the interest preference factor of the past,so that the user interest modeling is performed based on the prior interest of the user based on the interaction between the users.An improved algorithm based on user clustering is proposed to obtain the target user's neighbor collection more accurately.The recommendation system mainly predicts user's project scores based on the user's nearest neighbor set,so the accuracy of the nearest neighbor set is also very important for the recommended quality.In this paper,users are clustered by K-means algorithm,and then the nearest neighbor set of users is found out from the clustering of target users according to the similarity degree of users,so that the nearest neighbor set of users is closer to the target users.The trend of interest is more accurate.