Research On Collaborative Filtering Recommendat Ion Algorithm Combining User Rating And Attribute Interest

With the rapid advancement of science and technology and commercial applications,the Internet has entered every corner of daily life.While the popularity of the Internet has brought convenience to people's lives,research on how to use the vast data in the Internet to better serve people has never stopped.The recommendation system was born in the era of Internet data explosion.The recommendation system explores user interests by analyzing user attributes,user history behaviors,and attributes of recommended items,analyzes the characteristics of different users and different items,and proactively provides users with items that may be of interest to users.For the users,the recommendation system is a tool that actively provides the content in need,and can actively provide content that is of interest to the user without requiring additional operations;For merchants,the recommendation system is an effective content distribution tool that distributes the right content to the users who need it the most.When the number of common ratings of two users is too small,the similarity calculation will be difficult to obtain an accurate value.In this thesis,the birth and development of the recommendation system are introduced in detail.The basic principles and implementation process of the current popular recommendation algorithms are summarized,and the advantages and disadvantages of different recommendation algorithms are compared.Among them,the recommendation algorithm based on collaborative filtering is one of the most widely used recommendation algorithms in current research.It is introduced in detail in this thesis,and in the following content,the problems existing in the collaborative filtering recommendation algorithm are deeply studied and an improved algorithm is proposed.Constantly improving the recommendation accuracy and personalization is the pursuit goal of the recommendation system.Based on the in-depth analysis of the principle and process of the collaborative filtering recommendation algorithm,this thesis improves the problem of the similarity calculation in the traditional collaborative filtering recommendation algorithm.A collaborative filtering recommendation algorithm combining user rating and attribute interest "RACF" is proposed.First,judge the number of items that are commonly rated among users.When the number of common ratings of two users is too small,the traditional similarity calculation will be difficult to obtain accurate values.In this thesis,the user's rating on the project is mapped to the user's degree of interest in the item attribute,and the attribute interest feature vector is generated to calculate the user's similarity.When the number of two common ratings is sufficient,the rating habits of the two users are compared based on the degree of dispersion of the rating data between the users,and the similarity between the users is optimized based on the rating habit.Finally,simulation experiments on the Movielens-100k dataset show that the RACF algorithm improves the recommendation accuracy and the diversity of recommendation results.Although the RACF algorithm improves the diversity of recommendation accuracy and recommendation results,the user with the highest similarity is not necessarily the most helpful user for the recommendation.Therefore,based on the RACF algorithm,the generation of neighbor sets is further improved and optimized,and an improved collaborative filtering algorithm combining user rating and attribute interest "IRACF" is proposed.On the one hand,the algorithm considers that although some users have high similarity with the target users,the rated items are highly coincident with the items rated by the target users,and the recommended ability for the target users is weak.Therefore,this thesis proposes availability index,which reduces the probability that those users with high similarity but less help for recommendation become the nearest neighbor users through availability index.On the other hand,considering that users who are close to-1 in similarity to the target user and their rating of the item tend to present an opposite trend to the target user.Therefore,this thesis proposes the concept of negative neighbors,which uses negative neighbors to form a negative neighbor to predict the target user's ratings from the reverse side,and then gives the nearest neighbor and negative neighbor's prediction ratings different weights to get the final prediction rating.Finally,simulation experiments on the Movielens-100k dataset show that the IRACF algorithm further improves the diversity of recommendation accuracy and recommendation results.