Research On Recommendation Diversity Based On User Interest Depth

With the rapid development of Internet technology,the scale of users continues to expand,and the amount of data generated has increased rapidly,the user is surrounded by a sea of information every day.The recommendation system can filter information to improve the efficiency of information acquisition,and has received more and more attention and research.Traditional recommendation algorithms usually pay more attention to the accuracy of recommendation results.Blindly pursuing the accuracy of recommendation results can easily reduce its diversity.Therefore,how to effectively balance the accuracy and diversity of recommendation results is an important topic in the research of recommendation systems.This thesis has launched a research on this subject,and the main contents are as follows:1.From the perspective of the depth of user interest(the average number of historical interactions the user has in each field of interest),analyze the real data set.The analysis results show that users with greater depth of user interest are more likely to accept highly diverse recommendation results.On the contrary,users with lower interest depth are more likely to accept recommendations with high accuracy.Users of different interest depths have obvious differences in their diversity needs.2.We propose an improved algorithm d UCFQ based on collaborative filtering recommendation algorithm.This thesis first analyzes the characteristics of the original user's nearest neighbor collaborative filtering recommendation algorithm,and combines the user's interest depth to improve the user's nearest neighbor collaborative filtering algorithm to achieve different levels of recommendation according to the user's interest depth.service.Then this thesis conducted experiments on two real data sets.The experimental results show that d UCFQ can increase the diversity of recommendation results to a certain extent,and no significant decrease in accuracy has been observed.Finally,this thesis develops a movie recommendation system that collects user information from the web to obtain user viewing needs,generates a movie recommendation list and displays it to users.3.We propose a recommendation diversity algorithm model IRGAN-u M based on Generative Adversarial Networks.This thesis proposes a diversity algorithm model based on the generative adversarial network,matrix factorization technology and the maximum edge-related diversity algorithm that combines the depth of user interest to generate diverse and accurate recommendation results.Through experimental verification on two real data sets,the results show that the IRGAN-u M model can effectively improve the diversity of the recommendation results,and also performs well in the accuracy of the recommendation results.