A Recommender System Based on User Score Decomposition

Abstract

The invention discloses a recommendation system based on user score decomposition, including a data preprocessing module, a matrix decomposition module, a score prediction module, and a collaborative recommendation module, wherein the data preprocessing module obtains different user ratings for different commodities from the database of an e-commerce platform. The rating information of the user-product rating matrix is ​​constructed and stored; the matrix decomposition module obtains the weight value of the user rating decomposition through the simultaneous training of multiple models; the rating prediction module uses the weight value obtained by the matrix decomposition module to update the original rating matrix Decompose, obtain the predicted score value of each part by retraining, and finally store the sum of the predicted score values ​​of each part in the database as the score prediction result; the collaborative recommendation module uses the score prediction result calculated by the score prediction module Make recommendations to users. The present invention can quickly and accurately provide personalized recommendation services for users under the condition of less user history data.

Description

technical field [0001] The invention belongs to the field of recommender system research in the field of machine learning, and in particular relates to a recommender system for recommending by decomposing user ratings and mining user interest preferences. Background technique [0002] With the continuous development of the information industry and the rapid development of Internet technology, people have entered an era of information explosion. Faced with a huge amount of information every day, users cannot quickly find the information they want, that is, the phenomenon of information overload. [0003] It is precisely because of the rapid increase in the amount of information that led to the birth of the recommendation system. The recommendation algorithm in the recommendation system analyzes a large number of historical behaviors of users (such as browsing, clicking, scoring, and commenting), and finds the products they need from the vast amount of data for users, which g...

AI Technical Summary

Problems solved by technology

However, it still has the problem of over-fitting easily. Due to the excessive emphasis on the approximation of a single training score, the generalization ability of the model on the test set will be poor.

At the same time, the matrix factorization algorithm is poor in interpretability, and it is impossible to map each element of the user preference vector and commodity attribute vector to every purchase factor that affects users in real life.

Images