A Recommendation Algorithm Based on Multi-Relationship Network

Abstract

The invention discloses a multi-relational network-based recommendation algorithm. The multi-relational network-based recommendation algorithm comprises the following steps: (1) classifying a user data set according to different relational networks, and representing the relational networks as matrixes; (2) for users of each single-relational network, calculating a comprehensive link distance between two users; (3) calculating the Jaccard coefficient between the users according to the comprehensive link distances, between the two users, in various single-relational networks, obtained in the step (2); and (4) carrying out binarization processing according to the Jaccard coefficient between the users, building a comprehensive network and finally finishing the final recommendation task according to the single-relational network-based TrustWalker algorithm. According to the multi-relational network-based recommendation algorithm, the cold-start problem in a recommendation algorithm process is solved by combining the realities of the users and fully utilizing the information of the users in a plurality of social relational networks; and the final recommendation result is improved.

Description

technical field [0001] The invention relates to the technical field of information processing, in particular to a recommendation algorithm based on a multi-relationship network. Background technique [0002] There are many technical methods for building a recommendation system, among which the collaborative filtering recommendation algorithm is the most common and basic method currently used. But the traditional collaborative filtering algorithm encounters some problems, such as cold start problem, sparsity problem and so on. However, the research found that the social network that users participate in provides rich information for completing recommendation tasks, which can not only alleviate these problems to a certain extent, but also obtain better recommendation results. [0003] At present, there are many recommendation algorithms based on single-relational social networks, and experiments have also proved that the addition of social network information can help improve...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to overcome the deficiency that existing recommendation algorithms cannot consider multiple social relationships of users, and provide a recommendation algorithm based on multi-relationship networks, which can not only improve the effect of recommendation, but also improve the problem of cold start

Images