Optimization Of Recommendation Algorithms Based On Novelty And User Features

With the rapid development of the Internet,cloud computing,big data and other fields,the data on the network is growing exponentially.How to find out the content that users are interested in the huge ocean of data has become a difficult problem.Personalized recommendation system is one of the ways to solve these information overload problems.It can make personalized recommendations based on user characteristics and interest preferences.Collaborative filtering technology is one of the classic algorithms in the field of personalized recommendation.Collaborative filtering mostly uses user rating data for recommendation,and can adapt to various specific recommendation scenarios.However,its recommendation effect will be affected by data sparsity,cold start,and poor similarity calculation accuracy.Moreover,the current research does not consider the impact of novelty and user characteristics on recommendation.In this regard,this thesis proposes two hybrid recommendation algorithms that take into account the two factors of novelty and user characteristics.To solve the problems of data sparseness,poor recommendation accuracy and low novelty in collaborative filtering,the main work of this research is as follows:(1)Research and summarize the related knowledge of novelty recommendation and recommendation using user characteristics,etc.The common algorithm principle and solution process in the recommendation field are explained,and the current research status and challenges faced by the recommendation system are expounded,which provides a theoretical basis for the following research.(2)An optimized collaborative filtering algorithm that introduces novelty is proposed.The algorithm defines the calculation of the novelty degree,and combines the novelty degree and the Bias SVD algorithm to fill the sparse scoring matrix;uses the filled scoring matrix and user feature similarity to make the nearest neighbor recommendation.(3)A collaborative filtering algorithm that integrates user characteristics and ratings is proposed.This algorithm uses text information and user feature information to improve the nearest neighbor collaborative filtering algorithm.First use the TF-IDF algorithm to generate keywords for each movie,and then use the normalization method to generate user-keyword weight tables,gender-keyword weight tables,age-keyword weight tables,and occupation-keyword weight tables.Combining the weights of the three user characteristics of gender,age,and occupation with the weights of user keywords to obtain a user-keyword weight table that integrates user characteristics.Use the data in this weight table and the user’s score data to generate a user-keyword score table that combines user characteristics,and finally uses the user-keyword score data that fuses user features and the improved Jaccard similarity to perform nearest neighbor recommendation.Through repeated verification on the public data set,the novelty-introduced optimized collaborative filtering algorithm proposed in this paper can alleviate the problem of sparse scoring,and can get recommendations with higher novelty,which improves the situation of single recommendation results and poor freshness of users.The collaborative filtering algorithm that integrates user features and ratings proposed in this paper makes full use of user feature information and text information,making the recommendation results more in line with the real thoughts of users and improving the accuracy of the algorithm.