Multi-task recommendation method and system for joint Bayesian inference and weighted rejection sampling

Abstract

The disclosure provides a multi-task recommendation method and system for joint Bayesian reasoning and weighted rejection sampling. Among them, the multi-task recommendation method includes: constructing a Bayesian model embedding potential prior factors and a sampling model embedding balance factors running in parallel; determining the total number of recommended products and embedding Bayesian models embedding potential prior factors and embedding balance The sampling model of the factor is the ratio of the number of recommended products to the total number of recommended products; the corresponding negative sequence pattern mined from the specific user shopping behavior data set is input into the Bayesian model embedded with potential prior factors, and the probability Sequentially recommend a corresponding number of products to a specific user; at the same time, use the sampling model embedded in the balance factor to sample all user shopping behavior data sets, and recommend a corresponding number of products to a specific user.

Description

Technical field[0001]This disclosure belongs to the product recommendation area, in particular, involving a multi-task recommendation method and system for combined Bayesian reasoning and weighted denial sampling.Background technique[0002]The statement of this section is merely the background technology information related to the present disclosure, which is not necessarily constituted in prior art.[0003]For the field of user shopping behavior data, the sequence mode can indeed achieve good results for user behavior analysis, and can be interpreted. However, most sequence modes are only focused on the forward movement of the user. If a user purchases bananas, apples, but pears, for the user's negative action, such as the user's purchase of bananas, basketball, Apple's concern, showing more Low attention. However, in real life, the application and reference value of negative sequences is not less than positive sequence. When the user generated explicit feedback, negative sequence mod...

AI Technical Summary

Problems solved by technology

But at the same time, one thing that cannot be bypassed is that the neural network model is like a "black box". also a lot less

Although traditional methods such as collaborative filtering and singular value decomposition are highly operable and have certain interpretability, with the development of the times, they are increasingly unable to meet the individual needs of users, and it is necessary to innovate or propose new algorithms to replace

At present, based on Bayesian network reasoning, almost all focus on "forward reasoning", that is, to build graph reasoning through the behavior records generated by the user's explicit feedback, and rarely consider the user's potential preferences. The above problems make product recommendation Personalization bias, slow recommendation efficiency and inaccurate recommendation results

Images