Recommendation method and system based on heterogeneous information network and adaptive denoising

Abstract

The invention discloses a recommendation method and system based on a heterogeneous information network and adaptive denoising, and belongs to the technical field of recommendation systems. According to the method, the heterogeneous information network is adopted as the input of the model, complex input can be better processed, the real world can be better modeled, the problem that a homogeneous information network cannot simulate the real situation is solved, and the recommendation precision is greatly improved. And an adaptive denoising module is added, so that noise in the data set can be effectively removed, and effective data of the heterogeneous information network data set can be better reserved. And meanwhile, the depth map convolutional neural network is assisted, so that the precision of the recommendation model can be effectively improved, and the method adapts to a more complex real world. The problem that most of current recommendation systems only pay attention to recommendation precision is solved, the recommendation effect of the model can be changed from the perspective of the data set, meanwhile, due to the fact that the denoising module is added, the time complexity of the model becomes low, and the method can be effectively applied to the actual industrial environment.

Description

technical field [0001] The invention belongs to the technical field of recommendation systems, and relates to a recommendation method and system based on heterogeneous information networks and adaptive denoising. Background technique [0002] In recent years, recommender systems have played an increasingly important role in different online services, helping users discover items of interest in huge data sources. [0003] In the field of recommender systems, many previous studies are based on homogeneous information networks. One advantage of this is that modeling is more convenient, there is no need to consider different types of nodes and edges, and the implementation of the algorithm is more natural. [0004] However, the types and relationships of information networks in the real world are multi-faceted, and various forms of auxiliary information such as user, commodity, director, starring and other entity information can be used in the recommendation system. It is diff...

AI Technical Summary

Problems solved by technology

[0005] In view of this, the present invention provides a recommendation system based on heterogeneous information network and adaptive denoising, which solves the problem that most of the current recommendation systems only focus on recommendation accuracy, and only focus on recommendation accuracy without considering the diversity of data sets. For example, homogeneous graphs and heterogeneous graphs will lead to inability to adapt to more complex data sets, unable to better simulate the real world, and the obtained embedding will be greatly reduced

At the same time, compared with the existing heterogeneous information network recommendation system, the present invention designs a better meta-path model, and integrates a better self-adaptive denoising mechanism at the same time, which can obtain better recommendation effect and solve the current problem Some heterogeneous information network recommendation systems have low accuracy and cannot effectively extract embedding problems in datasets containing a lot of noise

Images