Research On Hybrid Recommendation System Based On Interest Stream

With the rapid development of the Internet,the increase of the information resource index has brought convenience to people's life,and it also brings people a lot of trouble in the choice of information.How to provide timely,accurate and personalized information for each user has become one of the most concerned issues in e-commerce.Under such a form and background,the theoretical research and practical application of recommendation system have been greatly developed.How to describe users' interest information and establish interest models has become a problem that must be solved by recommender systems.This paper makes a thorough research and Exploration on user's interest flow model and personalized recommendation system,and establishes a hybrid recommendation system model based on interest flow,and uses this model to make recommendations on different data sets.The main research contents of this dissertation include item-based time series network analysis,interest stream transition probability matrix analysis and construction,time window selection and time decay function strategy analysis,interest stream model recommendation result analysis,and mixed recommendation system modeling and analysis.In the entire recommendation process,it is divided into two major phases,namely the establishment phase of the sub-model and the hybrid model,and the stage of the recommendation based on the hybrid model.In the establishment phase of the sub-model,a detailed experimental verification of the proposed interest flow model was conducted.It was concluded that the model performed better than the existing models in some aspects,and the complexity of the model was better than the existing ones.The complexity of the model is low.In addition,the strategy of time window selection and time decay function is described in detail,and it is verified that the recommended effect is better under the combined effect of time window and time attenuation.In the multi-model fusion forecasting recommendation stage,each index of the sub-model is used as a weight coefficient,and a weighted sum is obtained for the recommended results generated by each sub-model to obtain a final mixed model recommendation result.The final indicator shows that the mixed recommendation based on the F-value index has the best overall effect.The experimental results of the mixed model show that the model mixing can combine the advantages of each model,and the shortcomings of each model can be made up to a certain extent,thus improving the recommendation quality of the whole recommendation system.