Optimization And Application Of Two-Step Recommendation Framework Based On Popularity Normalization And User Characteristics

Nowadays,with the increase of Internet use,massive data about People’s Daily life is generated,and the rational use of data is very important.Recommendation algorithm is an important technology for recommendation system to process data reasonably,and can also provide automatic recommendation for users.In order to attract more users to browse the application and extend the user browsing time,most applications are currently configured with recommendation system for promotion.These applications involve almost all aspects of the user,such as movies,music,books,travel,financial products and so on.Recommendation algorithm has gradually become an indispensable part of people’s network life.Recommendation algorithm is the technology of accurate recommendation to users.In order to solve the problem of inaccurate recommendation popularity proportion and similarity calculation,this paper makes improvements and optimization in the following aspects based on collaborative filtering algorithm and two-step recommendation framework.The specific contents are as follows:(1)Based on the traditional recommendation algorithm,the popularity weight factor is introduced to eliminate the influence of popular items on the recommendation results.First,the user-product scoring matrix is calculated.Then,taking the number of people interacting with the project as the index,the popularity of the project was defined and introduced into the scoring matrix.Finally,the similarity calculation function is improved to reduce the influence of popular products in the calculation of similarity and final recommendation.(2)A hybrid recommendation model was constructed considering user characteristics to solve the problem of cold startup.Combining demographics-based recommendation and collaborative filtering recommendation,a hybrid recommendation algorithm based on popularity and user characteristics(CPCF algorithm for short)is proposed.Specifically,the predictive score calculated based on user characteristics is linearly combined with the predictive score calculated by collaborative filtering,and the weight of both is measured by the precision of the recommendation algorithm.(3)In order to improve the recommendation accuracy,combined with the two-step recommendation method with excellent recommendation accuracy,a two-step recommendation algorithm based on popularity normalization and user characteristics(CPCF-TSP)was proposed.CPCF-TSP is to optimize each step of the two-step recommendation method.In the first step,user characteristics are introduced as a weight factor when predicting the probability of selecting rating items.In the second step,the mixed recommendation model CPCF proposed in research Content 2 is used to predict the scores of selected items,so as to complete the algorithm CPCf-TSp proposed in this paper.Under the framework of two-step recommendation method,CPCF-TSP can make full use of the demographic characteristics of users,and improve the similarity through popularity normalization,so as to help users choose suitable products.In addition,an example experiment is conducted using Movie Lens movie score data and Santander bank customer transaction data.The results show that the algorithm has excellent recommendation performance,and is especially suitable for the recommendation scenario of financial products where the user information is easy to collect and the order of user is much larger than the order of item.