Trust-based social collaborative filtering recommendation method

Abstract

The invention discloses a socialization collaborative filtering recommendation method based on trust. Two kinds of data which are a grading matrix of a user to a project and a user trust network are fused to carry out recommendation of high quality on an objective user, and the socialization collaborative filtering recommendation method based on trust mainly comprises the following steps that characteristic vectors of the user and the project are constructed; a grading prediction model Truster-MF facing user browsing behavior is built; a grading prediction model Trusteree-MF facing user evaluation behaviors is built; the model Truster-MF is trained; the model Trusteree-MF is trained; the trained Truster-MF and the trained Trusteree-MF are fused, and grading prediction models comprehensively considering the browsing behaviors and the evaluating behaviors is built. Compared with the prior art, the socialization collaborative filtering recommendation method based on trust mainly has the advantages that the socialization collaborative filtering recommendation method based on trust effectively solves the two major problems of data sparse and cold boot faced by collaborative filtering recommendation, has a better recommendation quality, and is compact and efficient in algorithm, easy to implement and particularly suitable for a large-sized electric commerce website.

Description

technical field [0001] The invention belongs to the field of information retrieval, in particular to a collaborative filtering recommendation method. Background technique [0002] The recommendation system can proactively push items of interest to users (such as news, books, movies, and music, etc.), and is an effective tool to solve Internet information overload. It has been widely used in various e-commerce websites and social networks, and has generated huge economic benefits. [0003] The core of the recommendation system is the recommendation algorithm. At present, there are many recommendation algorithms, mainly including collaborative filtering recommendation, content-based recommendation and hybrid recommendation. Among all recommendation algorithms, collaborative filtering algorithm is considered to be the most simple and effective, and has been successfully applied in many large-scale commercial recommendation systems. The basic principle of collaborative filter...

AI Technical Summary

Problems solved by technology

In addition, when constructing the matrix factorization model and algorithm, the above method emphasizes how to better fit the observed rating data, while ignoring the generation mechanism of the rating data, and does not consider how the observed user's evaluation of the item is based on generated by other users' reviews

Due to the above main reasons, the existing recommendation methods based on matrix decomposition have limited use of trust relationship data, the quality of recommendation is not improved, and they cannot solve the data sparse and cold start problems faced by collaborative filtering.

Images