Method and device for item recommendation

Abstract

The invention discloses an article recommendation method and device. The method comprises steps that according to all the pre-established users, initial articles of all the articles purchased by all the users and a resource distribution matrix of the users of historical data, a positive similarity matrix and a reverse similarity matrix are determined; an initial bidirectional similarity matrix is determined, and a second-order initial bidirectional similarity matrix is determined according to the initial bidirectional similarity matrix; a target deletion redundancy similarity parameter corresponding to the second-order initial bidirectional similarity matrix is determined; a target bidirectional similarity matrix is determined; all the users, target articles of all the articles purchased by all the users and the resource distribution matrix of the users are determined, and the target articles and recommendation articles corresponding to all the users of the resource distribution matrixes of the users are determined through the target articles and the resource distribution matrixes of the users. The method is advantaged in that more accurate user article recommendation is realized, and more customized recommendation is realized.

Description

technical field [0001] The present invention relates to the technical field of the Internet, in particular to an item recommendation method and device. Background technique [0002] With the rapid development of Internet technology, more and more useless information appears in the user's recommendation list. In a large number of shopping websites, how to recommend useful information for users and how to recommend more accurate information for each user, and then the item recommendation system appeared. [0003] Collaborative filtering method based on bipartite graph network is the most commonly used in item recommendation system. The main implementation method is to study the similarity of items based on the causal relationship between items, that is, to recommend items for users based on the similarity of items that users like successively in time sequence. In the actual process, the similarity between items is not based on the causal relationship in time, and it is not b...

AI Technical Summary

Problems solved by technology

[0005] In the existing technology, only positive similarity is considered, which may easily lead to biased item similarity estimation.

In addition, in the actual calculation process, the use of bidirectional similarity items has the problem of over-recommendation of redundant similarity items, that is, the weights of some recommended items obtained according to the bidirectional similarity matrix are the same, but in fact some Items are purchased by users based on a single attribute of user preferences, while other items may be purchased by users based on multiple different attributes of user preferences. Users must prefer items with multiple attributes that meet their preferences. The similarity of these items in the calculation results The weights are the same, but users have different preferences for them. As a result, some items purchased by users based on a single attribute of user preferences are overestimated by the system and recommended to users, instead of highlighting items that meet user needs. Accuracy of Recommended Items

Images