Item recommendation method and item recommendation device

Abstract

The embodiment of the invention provides an item recommendation method, which comprises the following steps that: N preference values are determined, wherein each preference value in the N preference values indicates the selection preference degree of a target user in a plurality of users on a target item in a plurality of items under the condition of X context type indications; a preference value tensor is determined according to the N preference values, and (X+2) matrix factors are determined according to the preference value tensor; and the item recommendation is carried out according to the at least one matrix factor in the (X+2) matrix factors. According to the embodiment of the invention, a plurality of matrix factors approaching to the tensor can be determined according to data sparseness preference value tensor values, and then, the item recommendation can be carried out according to the at least one matrix factor. The order of each matrix factor in a plurality of matrix factors is respectively smaller than the order of the initial preference value tensor, so that during the context recommendation, the recommendation accuracy can be ensured, and the calculation complexity is also effectively reduced.

Description

Technical field [0001] The present invention relates to the field of information technology, and more specifically, to a method and device for recommending items. Background technique [0002] The recommendation system establishes a binary relationship between users and items, utilizes the existing selection process or similarity relationship to dig out objects of potential interest for each user, and then makes personalized recommendations. [0003] In recent years, Adomavicius and Tuzhilin et al. pointed out that integrating contextual information into recommender systems will help improve the accuracy of recommendations, and proposed the widely cited concept of context-aware recommender systems (CARS). Context information such as It can be the time, location, mood and other contextual conditions when the user selects the item. [0004] The prior art uses a multi-dimensional vector model to represent context information for recommendation. Specifically, in context, Pearson's corr...

AI Technical Summary

Problems solved by technology

[0005] However, too many context-dependent dimensions will cause data sparsity, in other words, among all users, only a few users have preferences for the same item

Therefore, in the case of sparse data, calculations using the Pearson correlation coefficient usually cannot obtain high accuracy, which in turn affects the accuracy of the recommendation. If the accuracy is to be improved, a very complicated calculation process is required.

Images