Research And Application Of Recommendation Algorithm Based On Diversity Of User Preference

In recent years,with the spread of Internet technologies,services related to person-alized recommendations have been widely applied to all aspects of life.Personalized recommendation technology can not only help users find the desired content in massive information,but also can bring huge economic benefits to the merchant.The recom-mendation system can understand the user's interest preferences by analyzing the user's historical behavior,thereby providing the user with accurate recommendation content.The main content of this article is to analyze and mine the user's preference information from explicit(user's comments)and implicit(user's browsing behavior),so as to rec-ommend the merchant to the user and recommend the new customer to the merchant.How to fuse the user's preference information into the recommendation model,as well as the sparseness and heterogeneity of the data itself,present a great challenge to the rec-ommendation system.So this paper proposes corresponding models to solve the above challenges,specifically,including the following two aspects.Firstly,this paper proposes a merchant recommendation model based on multi-preference information of users for the problem of explicit user preference research.The user's review information contains user's preference information for different aspects.The integration of such information through effective modeling and algorithm design can greatly enhance the effectiveness of personalized recommendations.Specifically,this paper proposes a probability graph model that fuses user preference information to collaborative filtering.Then this paper uses a Gaussian model-based optimization method to train the parameters of the model.Finally,this paper verifies the validity and scalability of the model on two typical data sets.The experimental results show that the probabilistic graph model proposed in this paper is superior to the traditional methods in terms of mean absolute error(MAE),root mean square error(RMSE)and recall rate(RECALL),and performs better than the baseline method in cold start problem.Secondly,this paper constructs a new guest recommendation model that integrates user browsing and purchasing information for the problem of user implicit preference research.In the field of e-commerce,it is of great significance for users to pull new ones for business promotion and increase brand influence.The user's historical brows-ing information and purchasing information in the station largely reflect the user's in-terest preference.The integration of such information through reasonable modeling and effective algorithms can improve the effectiveness of the new guest recommendation.Specifically,this paper proposes a multilayer neural network recommendation model that integrates merchandise embedding information.Firstly,embedding the attributes corresponding to commodities such as brands and shops based on the historical brows-ing commodity sequence of the user.Then,each user is represented as a multidimen-sional vector in combination with the user's user profile information and browsing and purchasing information,and then input to a multi-layer neural network model.Finally,this paper verifies the new user pull effect of the model on two real electrical appliances merchants of an e-commerce company.The experimental results show that our method outperforms the traditional method in terms of AUC(Area Under Curve).