Research On Diversified Recommendation Algorithm Based On User Behavior Analysis

With the development of Internet technology,the rapid rise of e-commerce platforms and online social platforms has made the interaction between users and the outside world increasingly frequent.Recommendation systems can provide personalized content to users based on their historical behavior and interests.Existing recommendation systems not only need to consider users’ interests but also the diversity and differentiation of the contents,and provide users with more rich and diversified services and improve user satisfaction.Diversified recommendation algorithms mainly perform diversified product recommendation by studying the diverse interests of users.Currently,diversified recommendation algorithms mainly focus on reordering lists and optimizing objective functions to improve the diversity of recommendation results.However,existing methods ignore differentiated personalized interests of individual users and fail to distinguish similar interests among groups of users,and lacks the research of multi-level categories of user interests.This paper introduces the difference of individual characteristics information and group characteristics information of users into the recommendation algorithm,and a matrix factorization recommendation algorithm incorporating diversity regularization terms and a diversity collaborative filtering recommendation algorithm by integrating self-organizing maps neural network are proposed.The main research content are as follows:(1)By considering the individual feature information of users,a matrix factorization algorithm by integrating with diversity regularization terms is proposed.Firstly,according to the user information,product information and interaction records between users and products,the category difference degree between products and the degree of user interest diversity are calculated.The diversity regularization factor is constructed by using the category difference score and the user interest diversity score to adjust the user’s score for different categories of products.Then,the user’s latent vector matrix is learned through the degree of user interest diversity,and the latent factor representation of each product is modeled by the difference degree between the categories of products,which can form the latent vector matrix of products.Finally,the latent vector matrix of users and products are used to predict the ratings of users for products.The experiments show that the matrix factorization algorithm with the introduction of diversity regularization factors improves CC@10 and ILD@10 by 10.8% and 7.2% on the Douban movie dataset,and improves 7.9%and 6% on the Book dataset,respectively,compared to traditional algorithms.The method can effectively improve the diversity of recommendation results and enhance user satisfaction.(2)By considering the group characteristics of the users,a diversity collaborative filtering recommendation algorithm by integrating self-organizing maps neural network is proposed.Firstly,a user-product and user-product category rating table are constructed based on the user ratings of products.Then,the collaborative filtering algorithm is used to obtain a product recommendation list based on the users with similar ratings.Secondly,the user vectors are input into a self-organizing maps neural network to cluster similar user groups.The similar user groups are used to find the product categories that the target user may be interested in,which can form a diversified recommendation list.Finally,the two recommendation lists are integrated to form a product recommendation result that satisfies the diversity and accuracy.Experiments show that the similar user group based on selforganizing maps neural network can enhance the interest diversity of the target user and effectively improve the diversity of recommendation results.Compared with traditional algorithms,the CC@10 and ILD@10 indicators improve by 1.6% and 1.9% on the Amazon Music dataset and 1.3% and 1.4% on the Beauty dataset.