DE-112023006663-T5 - Method and setup for recommendations based on a neural graph network

DE112023006663T5DE 112023006663 T5DE112023006663 T5DE 112023006663T5DE-112023006663-T5

Abstract

A method for generating recommendations based on a neural graph network is disclosed. The method includes receiving a bipartite user-element graph consisting of a set of user nodes representing a set of users, a set of element nodes representing a set of elements, and a set of edges representing interactions between the users and the elements; obtaining user embeddings for each of the set of users and element embeddings for each of the set of elements; performing an aggregation process within layers to generate aggregated user embeddings for each of the set of users and aggregated element embeddings for each of the set of elements based on different weighting coefficients for each user and each element; and performing a propagation process between layers to generate propagated user embeddings for each of the set of users and propagated element embeddings for each of the set of elements based on different weighting coefficients for each propagation layer. and making a recommendation for a user from the set of users based on the propagated user embeddings and the propagated element embeddings.

Inventors

KHARLAMOV EVGENY
TANG JIE
DONG YUXIAO
ZHANG DAN
ZHU YIFAN

Assignees

BOSCH GMBH ROBERT
UNIV TSINGHUA

Dates

Publication Date: 20260507
Application Date: 20230928
Priority Date: 20230928

Claims (15)

A method for generating recommendations based on a neural graph network, comprising: Receiving a bipartite user-element graph consisting of a set of user nodes representing a set of users, a set of element nodes representing a set of elements, and a set of edges representing interactions between users and elements; Receiving user embeddings for each of the set of users and element embeddings for each of the set of elements; Performing an inner-layer aggregation process to generate aggregated user embeddings for each of the set of users and aggregated element embeddings for each of the set of elements based on different weighting coefficients for each user and each element; Performing an inter-layer propagation process to generate propagated user embeddings for each of the set of users and propagated element embeddings for each of the set of elements based on different weighting coefficients for each propagation layer; and providing a recommendation for a user from the set of users based on the propagated user embeddings and the propagated element embeddings.
Procedure according to Claim 1 , where the interactions include one or more of purchase, rating, clicking and collection.
Procedure according to Claim 1 , where the weighting coefficients are calculated using a heat core method or a personalized PageRank method.
Procedure according to Claim 1 , where the weighting coefficients are calculated based on an input level of each user node or element node in an initial layer.
Procedure according to Claim 1 , furthermore comprehensively: Combining the propagated user embeddings and the propagated element embeddings of each propagation layer to generate final embeddings for each user and each element.
Procedure according to Claim 5 , furthermore comprehensively: Optimizing the neural graph network based on the final embeddings of each user and each element using a Bayesian personalized ranking loss (BPR loss).
A neural graph network-based recommendation system comprising: an input module for receiving a bipartite user-element graph consisting of a set of user nodes representing a set of users, a set of element nodes representing a set of elements, and a set of edges representing interactions between users and elements; an embedding module for obtaining user embeddings for each of the set of users and element embeddings for each of the set of elements; an aggregation module for performing an inner-layer aggregation process to generate aggregated user embeddings for each of the set of users and aggregated element embeddings for each of the set of elements based on different weighting coefficients for each user and each element; a propagation module to perform a propagation process between layers to generate propagated user embeds for each of the set of users and propagated element embeds for each of the set of elements based on different weighting coefficients for each propagation layer; and a recommendation module to make a recommendation for a user of the set of users based on the propagated user embeds and the propagated element embeds.
establishment according Claim 7 , where the interactions include one or more of purchase, rating, clicking and collection.
establishment according Claim 7 , where the weighting coefficients are calculated using a heat core method or a personalized PageRank method.
establishment according Claim 7 , where the weighting coefficients are calculated based on an input level of each user node or element node in an initial layer.
establishment according Claim 7 , furthermore comprehensive: a pooling module to combine the propagated user embeddings and the propagated element embeddings of each propagation layer to generate final embeddings for each user and each element.
establishment according Claim 11 , furthermore comprehensively: an optimization module for optimizing the neural graph network based on the final embeddings of each user and each element using a Bayesian personalized ranking loss (BPR loss).
Device for recommendations based on a neural graph network, comprising: a memory; and at least one processor coupled to the memory and designed to execute the procedure according to one of the Claims 1 - 6 to carry out.
A computer-readable medium that stores computer code for recommendations based on a neural graph network, wherein the computer code, when executed by a processor, causes the processor to perform the procedure according to one of the Claims 1 - 6 to carry out.
Computer program product for recommendations based on a neural graph network, comprising: processor-executable computer code for carrying out the procedure according to one of the Claims 1 - 6 .

Description

Technical field The present disclosure relates generally to technology with artificial intelligence and in particular to a recommendation system based on neural graph networks. background In the real world, some data structures can be represented by a graph. A graph can consist of nodes and edges. A node can be a person, a place, a thing, etc., and an edge can define the relationship between nodes. Graph neural networks (GNNs) are special types of neural networks capable of working with a graph data structure. GNNs can be used for predicting nodes, edges, and graph-based tasks. An input graph can be passed through a series of neural networks and transformed into a graph embedding, effectively preserving information about the nodes, edges, and global context of the input graph. In recent years, GNNs have made significant progress in recommendation tasks. The core mechanism of GNN-based recommendation systems is the iterative aggregation of neighboring information on the user-element interaction graph. However, existing GNNs treat users and elements equally and cannot distinguish different local patterns of each node, making them suboptimal for recommendation scenarios. Therefore, the challenge lies in improving GNN-based methods for recommendations. Brief description The following is a simplified summary of one or more aspects according to the present disclosure, in order to provide a basic understanding of these aspects. This summary is not a comprehensive overview of all aspects under consideration and is not intended to identify key or critical elements of all aspects, nor to outline the scope of any one or all aspects. Its sole purpose is to present some concepts of one or more aspects in simplified form as an introduction to the detailed description that will be presented later. In one aspect of disclosure, a method for making recommendations based on a neural graph network is disclosed. The method includes receiving a bipartite user-element graph consisting of a set of user nodes representing a set of users, a set of element nodes representing a set of elements, and a set of edges representing interactions between the users and the elements; obtaining user embeddings for each of the set of users and element embeddings for each of the set of elements; performing an aggregation process within layers to generate aggregated user embeddings for each of the set of users and aggregated element embeddings for each of the set of elements based on different weight coefficients for each user and each element; and performing a propagation process between layers to generate propagated user embeddings for each of the set of users and propagated element embeddings for each of the set of elements based on different weight coefficients for each propagation layer. and making a recommendation for a user from the set of users based on the propagated user embeddings and the propagated element embeddings. In another aspect of the disclosure, a device for recommendations based on a neural graph network is disclosed. The device includes an input module for receiving a bipartite user-element graph consisting of a set of user nodes representing a set of users, a set of element nodes representing a set of elements, and a set of edges representing interactions between the users and the elements; an embedding module for obtaining user embeddings for each of the set of users and element embeddings for each of the set of elements; an aggregation module for performing an inner-layer aggregation process to generate aggregated user embeddings for each of the set of users and aggregated element embeddings for each of the set of elements based on different weighting coefficients for each user and each element; a propa A propagation module to perform a propagation process between layers to generate propagated user embeds for each of the set of users and propagated element embeds for each of the set of elements based on different weighting coefficients for each propagation layer; and a recommendation module to make a recommendation for a user of the set of users based on the propagated user embeds and the propagated element embeds. In another aspect of the disclosure, a device for computer vision processing is disclosed. The device may include a memory and at least one processor coupled to the memory. The at least one processor may be configured to: receive a bipartite user-element graph consisting of a set of user nodes representing a set of users, a set of element nodes representing a set of elements, and a set of edges representing interactions between the users and the elements; obtain user embeddings for each of the set of users and element embeddings for each of the set of elements; perform an inner-layer aggregation process to generate aggregated user embeddings for each of the set of users and aggregated element embeddings for each of the set of elements based on different weighting coefficients for each user and each element; Perf