Topic model analysis and user short-long interest-based recommendation method

Abstract

The invention discloses a topic model analysis and user short-long interest-based recommendation method. The method comprises the steps of performing hidden topic training on an activity text set to obtain a trained model, and performing calculation to obtain type characteristic topic distribution of all types; inputting short-term and long-term activity texts of a user to the trained model to obtain short-term and long-term activity topic eigenvectors of the user, and in combination with behavior weights and a time decay function, obtaining short and long interest vectors of the user; and according to the long interest vector of the user and the type characteristic topic distribution of all the types, obtaining a cosine similarity value of a long interest, selecting the type of TopM withthe highest cosine similarity value of the long interest, according to the short interest vector of the user and activities in the type of the TopM, obtaining a cosine similarity value of the short interest, and according to the cosine similarity value of the short interest, obtaining recommended activities of an activity platform for the user. The screening range is reduced when the to-be-recommended activities are selected, so that the recommendation calculation time is shortened and the recommendation accuracy can be improved.

Description

technical field [0001] The invention belongs to the field of recommendation algorithms in recommendation systems, and more specifically relates to a recommendation method based on theme model analysis and users' long and short interests. Background technique [0002] In recent years, activity-based social networks (EBSNs), such as meetup and Douban.com, have been extensively developed. These websites not only provide a convenient platform for disseminating various social activities, but also build a huge social network among users. As a result, whether in academia or industry, how to efficiently recommend personalized activities for users has become a hot field. Activity recommendation is different from other product recommendations, because product recommendations are usually not embedded in social networks, and various complex social relationships are not considered. Therefore, activity recommendation in social networks faces many new challenges. [0003] When using the...

AI Technical Summary

Problems solved by technology

[0005] In view of the above defects or improvement needs of the prior art, the present invention provides a recommendation method based on topic model analysis and user long-short interest, thereby solving the problem that the prior art does not consider that the user's long-short interest may change over time, The technical problem of not distinguishing between different behavior categories in user behavior records, nor adding behavior weight factors for weighting considerations, resulting in inaccurate recommendation results

Images