Research On Personalized Recommendation Algorithms Based On Hybrid Model

Along with the development of Internet and Information technology,the phenomenon of information overload is becoming increasingly significant.As one of the most effective ways to solve the problem,the recommender systems has drawn a great attention since it appeared,and been widely used in various fields,such as product recommendation in e-commerce,books,movies or music recommendation in multimedia website,personalized advertisement recommendation etc.The recommender systems have created tremendous commercial value that with good development and application prospects.Nowadays,most of the mainstream recommendation algorithms use user behaviors to capture the user's interest in the items.There are some problems in this way,user behavior represents a kind of habit of users and only contains positive feedback information,which is not sufficient to accurately reflect user preference.Besides,traditional recommendation algorithm is limited to single data source and its framework,which can not overcome its own shortcomings and is vulnerable to targeted attacks,attackers can take advantage of characteristics of the algorithm to induce the system to recommend specific items.In this paper,in order to solve these problems,we utilize sentiment analysis technology to analyze user comments,and propose a hybrid preference model that considers both user behavior preference and sentiment preference for user preference mining,which is used for recommendation algorithms input as user initial preference.Moreover,the data source is fully utilized to improve the collaborative filtering algorithm,the matrix factorization algorithm and the graph-based diffusion algorithm.Based on the three improved algorithms,a new hybrid recommendation model is proposed to further improve the recommendation effect and enhance the system reliability.We conduct experiments on Zhejiang Migu reading datasets to verify the proposed hybrid model's performance in book recommendation system.The results show that proposed improved algorithm improves the accuracy of recommendation,meanwhile,the hybrid recommendation model can produce better performance than the independent ones,providing more accurate,higher novel recommendations.