Personalized recommendation method and apparatus used for sparse big data

Abstract

The invention discloses a personalized recommendation method and apparatus used for sparse big data. Behavior records generated between users and commodities can be obtained through a user historical behavior database, so that related data can be efficiently and comprehensively found, and a behavior matrix between the users and the commodities is generated; when the behavior records generated between the users and the commodities are relatively sparse, all the commodities in the behavior matrix are included in corresponding commodity clusters of a commodity cluster set through the similarity between the commodities, and the membership degrees of the users to the commodity clusters are calculated, so that the membership degrees can be used for describing the users; the membership degrees of the users to the commodity clusters can enable the characteristics of the users to be more remarkable; the similarity of the users calculated based on the membership degrees is more accurate; and the accuracy of recommendation based on similar users in collaborative filtering is improved. The commodity cluster dimension of a membership degree matrix is far smaller than the dimension of the commodities in the behavior matrix, so that the time and space resources of user similarity calculation are greatly saved and the recommendation efficiency is improved.

Description

technical field [0001] The invention relates to the field of computer technology, in particular to a personalized recommendation method and device for sparse big data. Background technique [0002] With the rapid development of the Internet, many personalized recommendation services have begun to appear. These personalized recommendation services can recommend to users information that is most likely to meet their interests based on the user's historical behavior records. [0003] In the prior art, there is a personalized recommendation method of collaborative filtering. This method considers that the user's interest is unchanged for a period of time, and can recommend to the user products that are similar to the user's interest. Therefore, the usual recommended The process includes two steps of similarity calculation and recommendation generation. The user's historical behavior is used as a feature to characterize the user, and then recommendations are made based on the si...

AI Technical Summary

Problems solved by technology

The distribution of user behavior is very uneven, usually following a power-law distribution or Weibull distribution. Most users have only a small amount of behavior. For a single user, the content with corresponding behavior is very sparse compared to all content. When user behavior is sparse, There are often few or no common behaviors between users. The recommended data can only describe and understand a user from a small number of dimensions, and it is difficult to fully judge the user's attributes, consumption level and ho

Images