Research On Recommendation Algorithm Based On Topic Model And User Similarity

Since entering the 21 st century,computer technology has made rapid development.At present,people's work,study,information acquisition,and many other aspects are closely related to the Internet.While the popularization of the Internet makes life more convenient,it also brings problems such as information overload.A recommendation algorithm can effectively alleviate this problem and can recommend the information that users are interested in.In this thesis,a recommendation algorithm based on the topic model and user similarity is proposed by considering the interests and preferences of individual users and the factors influencing each other.The main work includes:(1)A recommendation algorithm based on topic model and user similarity is proposed.The algorithm uses LDA(Latent Dirichlet Allocation)topic document generation model to generate the probability matrix of user preferences for items.Clustering users according to their characteristics and calculating the weights among users.The user preference information is fused with the user's weight,and the user's final preference is calculated to produce the recommendation result.Compared with the existing recommendation algorithms,the proposed algorithm improves the recall rate,precision rate,and other indicators,and reduces the difficulty of solving the problem of data sparsity.(2)A method to calculate the weight between users by grouping information is proposed.Methods Clustering the users according to the data of user tags,dividing the users with similar preferences into the same group,and calculating the weight among users according to the information of user groups.The user preferences will change over time.The method calculates the time-based user similarity according to the time when the user marks the item.Methods The weight of user group information and similarity information were fused to get the final weight among users.(3)The method of mining user interest based on labels is put forward.Methods User-label data and label-item data were processed by using the LDA topic model,and semantic information in tags was integrated into the binary data of user and item,to obtain the probability of user preference for an item.Methods The weight of users and their preference information were fused to simulate the process of user interaction,and the final preference of users was obtained.Compared with the traditional method which only considers the user's rating information and the number of tags,this method can better tap the potential interest of the user.In this thesis,the Last.fm data set commonly used in the field of recommendation algorithms is used for experiments,and the singer is recommended for users according to the music genres they are interested in.The experimental results show that the recommendation algorithm proposed in this thesis has a good effect on the music field and can optimize the user experience.