Spectrum collaborative filtering recommendation system and establishment method thereof

Abstract

The invention provides a spectrum collaborative filtering recommendation system and an establishment method thereof. The establishment method of the spectrum collaborative filtering recommendation system mainly comprises the following steps: S1, acquiring a data set of a user and a project; S2, preprocessing the data set to obtain a bipartite graph; S3, carrying out partial attribute weighting onthe bipartite graph, and carrying out graph Fourier transform to obtain a new bipartite graph; S4, placing a spectral convolution filter on the new bipartite graph to obtain a new graphic signal; S5,improving the spectral convolution filter by adopting a polynomial approximation method to obtain a final spectral convolution operation; and S6, establishing a recommendation system by adopting approximate convolution operation. Through a spectrogram theory, connection information between a user and a project is converted into a spectral domain, a recommendation system of multi-layer convolutionoperation is established, the problem of cold start in a common recommendation system is relieved, and the accuracy of the recommendation system is improved.

Description

technical field [0001] The invention relates to the technical field of recommendation systems, in particular to a spectral collaborative filtering recommendation system and its establishment method. Background technique [0002] With the birth and rapid development of network technology, the relationship between people and the Internet is becoming more and more inseparable. Various websites contain a large amount of disorganized information, and it is very difficult for users to quickly and effectively identify the information that is useful or interesting to them from the huge amount of information. [0003] In the context of big data in today's society, the research on recommender systems is of great significance. The recommendation system is a method to effectively recommend information that may be of interest to users based on the information they have browsed. The effectiveness of recommender systems often depends on how the user's interests or preferences are underst...

AI Technical Summary

Problems solved by technology

[0006] Therefore, the existing recommendation systems based on collaborative filtering or memory and models cannot meet the needs of users to select items with high reliability from a large number of unbrowsed items to recommend to users.

Images