Personalized recommendation method, system and storage medium based on discrete factorization machine

Abstract

The invention discloses a discrete decomposer-based personalized recommendation method, a discrete decomposer-based personalized recommendation system and a storage medium. The method comprises the steps of inputting historical transaction data of a user for the same category of commodities into a pre-established collaborative filtering model, and training the collaborative filtering model to obtain a trained collaborative filtering model; for all of commodity data, generating a first candidate set by use of the trained collaborative filtering model; inputting the historical transaction data into a discrete decomposer DFM (Dimensional Fact Model), and training the discrete decomposer DFM to obtain a trained discrete decomposer DFM; filtering the first candidate set by use of the trained discrete decomposer DFM to obtain second candidate sets; training a feature-based recommendation model by use of the historical transaction data to obtain a trained feature-based recommendation model, and sorting the second candidate sets by use of the trained feature-based recommendation model; and selecting the preset number of commodities which are ranked at the top from a sorting result as a final recommendation result.

Description

technical field [0001] The invention relates to a personalized recommendation method, system and storage medium based on a discrete decomposition machine. Background technique [0002] Personalized recommendations are ubiquitous in today's online world—almost every online activity can be considered a recommendation, such as news or music feeds, car or restaurant reservations, and online shopping. Therefore, an accurate recommender system is crucial not only to the quality of service but also to the profit of the service provider. In today's e-commerce portals, there are millions of products, but only a small part of them can attract the attention of users. For this reason, the industry usually divides personalized recommendation into two stages: the generation stage of candidate sets and the arrangement stage of candidate sets. For example, the generation stage of the candidate set is to select 1000 products from 1 million products as the candidate set for individual users...

AI Technical Summary

Problems solved by technology

[0004] However, the current personalized recommendation in the industry cannot strictly guarantee the accuracy of the candidate set generation process.

A bad situation is that due to the poor accuracy of traditional algorithms, many products with high user preferences fail to enter the candidate set in the first stage. Even if the performance of the recommendation algorithm in the second stage is good, the final recommendation result are very bad

Images