A Rating Prediction Method Based on Multi-source User Reviews

Abstract

The invention provides a rating prediction method based on multi-source user comments, which belongs to the field of recommendation systems. Perform data preprocessing on the user's historical consumption records; calculate the similarity between different users according to the user-item scoring matrix and the similarity formula, and use the relevant comments written by similar users with the highest similarity to supplement user comments; then extract comments features; finally, review feature fusion processing is performed. The present invention can filter out relevant comments written by similar users based on user historical consumption records, supplement user comments, and construct a user comment supplementary document for each user, which can alleviate the data sparse problem of user comments, enrich user portraits, and improve ratings The accuracy of the prediction can improve the user satisfaction with the recommendation system. In addition, user review supplementary documents are composed of related reviews written by similar users, which are different from reviews written by users themselves, which can improve the diversity of recommendations.

Description

technical field [0001] The invention belongs to the field of recommendation systems, in particular to a rating prediction method based on multi-source user comments. Background technique [0002] In today's Internet information overload situation, information consumers without clear needs want to find interesting content conveniently, and information producers want to push their content to suitable target users, and recommendation systems have emerged as the times require. Score prediction is a classic model in the recommendation system. The system predicts the score of all items that the user has not consumed, and then recommends the N items with the highest predicted scores to the user. At present, the most widely used method in the rating prediction problem is the collaborative filtering algorithm. The collaborative filtering algorithm uses user behavior data to mine the interests of users and recommend items that may be of interest to them. Classical work includes Matrix...

AI Technical Summary

Problems solved by technology

However, since the vast majority of users are unwilling to write comments to share their consumption experience, even if some users write comments, most of them are relatively short, so the user comments themselves are sparse, and the comments of most users cannot fully represent a user’s like

Images