Article cold start recommendation algorithm integrating relationship mining and collaborative filtering

Abstract

The invention discloses an article cold start recommendation algorithm integrating relationship mining and collaborative filtering. The algorithm comprises: first, using an item attribute matrix as abasis, calculating a plurality of binary relationships between every two attributes by adopting a relationship mining method; expanding limited article attributes into more relationship attributes; and then obtaining an attribute relation matrix, calculating the attribute similarity between the articles, meanwhile, fusing article scoring information for similarity weighting calculation, achievingpersonalized recommendation of the new articles. The problem of cold start of the new articles in a recommendation system can be systematically solved, and the recommendation accuracy and the articlediversity are improved.

Description

technical field [0001] The invention relates to the technical field of recommendation, in particular to an item cold-start recommendation algorithm for fusion relationship mining and collaborative filtering that can improve recommendation accuracy and item diversity. Background technique [0002] Recommender Systems (RS) are mainly used to predict the ratings of target users on other unrated items, analyze the user's preferences according to the target user's historical preference data, and recommend their favorite items for users. Existing recommendation methods can be divided into collaborative filtering (Collaborative Filtering, CF), content-based filtering (Content Based, CB) and hybrid methods. [0003] CF recommendation mainly uses user rating data to construct a rating matrix, calculates the similarity between users or items, predicts items that have not been rated by users, and recommends target users based on predicted ratings. CF relies on the relationship between...

AI Technical Summary

Problems solved by technology

This algorithm does not rely on the user's rating information on the item, but focuses on the content information of the item. Although it can effectively solve CS, it only recommends similar items based on the items that the user likes in the history, and the recommendation result is not ideal.

[0007] At present, many researchers have used hybrid filtering methods to solve CS problems, but due to the lack of item attributes and difficult feature extraction, the quality of recommendation is not high

In addition, researchers have also solved CS problems by introducing additional information, such as social information, item content descriptions, and comment information. Although the recommendation results have improved to a certain extent, it is difficult to obtain information, ignoring the complexity of interpersonal relationships and The real-time nature of the network environment cannot provide reliable resources for social recommendation in real time, nor can it solve the problems of CCS and ICS at the same time

Images