Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

Recommendation method based on multi-behavior session graph fusion

A recommendation method and behavioral technology, applied in marketing, advertising, instruments, etc., can solve the problems of reducing the diversity of recommendation results, ignoring users' future behaviors, and difficult for models to acquire "novel" items, so as to improve diversity and The effect of accuracy

Active Publication Date: 2021-12-31
山东省人工智能研究院 +2
View PDF9 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0003] 1. Only use a single type of user behavior data for training, ignoring that the user's future behavior is determined by multiple historical interaction behaviors;
[0004] 2. Only pay attention to the behavioral characteristics of a single user, ignoring the internal connection of behavioral patterns between similar users;
[0005] 3. In the same behavioral session, user interaction items are highly homogeneous, and it is difficult for the model to acquire "novel" items, which reduces the diversity of recommendation results

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Recommendation method based on multi-behavior session graph fusion

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0034] In order to ensure high connectivity and item diversity of the sparse behavior session graph, preferably, the value of N in step b) is 10.

Embodiment 2

[0036] Step c) comprises the following steps:

[0037] c-1) Take joint session itemsets Items in the adjacency matrix A are initialized as graph nodes click , to traverse the click sequence of the target user and neighbor users, if item j is the next item clicked by the user after item i is clicked, then construct an edge between item i and item j, that is, the adjacency matrix A click The middle abscissa is the value of i and the ordinate is j Increase by 1, for the adjacency matrix A click Normalize to get the click session graph

[0038] c-2) Take joint session itemsets Items in the graph node initialize the collection adjacency matrix A collect , to traverse the collection sequence of the target user and neighbor users, if item j is the next favorite item after the user’s favorite item i, then construct an edge between item i and item j, that is, the adjacency matrix A collect The middle abscissa is the value of i and the ordinate is j Increase by 1, for the a...

Embodiment 3

[0041] In step d) through the formula Calculate the final embedding vector e obtained by the high-order propagation of act behavior i,act , act is one of the behaviors of click, collection and purchase, is the embedding vector of message aggregation of node i layer l in the act behavior session graph, where a l is the propagation weight, L is the total number of layers of propagation, L=2, is the l-layer neighbor node item set of node i in the act behavior session graph, for e i,act The weights of node i and neighbor node j in the corresponding adjacency matrix, for The embedding vector of node j in .

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The invention discloses a recommendation method based on multi-behavior session graph fusion. The method comprises the steps: constructing a multi-behavior weighted undirected session graph by using joint multi-behavior sequence data of a target user and similar users; on the basis, acquiring project multi-behavior embedding according to different weight aggregated neighbor information, and acquiring user interest characterization by serially connecting project multi-behavior embedding and combining an attention mechanism; and finally, performing inner product by using project embedding and user interest characterization to obtain a normalized score to decide whether to recommend the project or not. Compared with other session type recommendation methods, the method has the advantages that firstly, project embedding containing more behavior intentions can be obtained from modeling user multi-behavior sequence data; secondly, the sequence is constructed into a weighted undirected graph, one-way constraint during neighbor information aggregation is relieved, and the model can learn the two-way relation between items; thirdly, similar users are used for supplementing target user data, and the model can learn'novel 'items which do not appear in target user historical data, so that the diversity and accuracy of recommendation results are improved.

Description

technical field [0001] The invention relates to the technical field of conversation graph recommendation, in particular to a recommendation method based on fusion of multi-behavior conversation graphs. Background technique [0002] Effectively capturing user interests is the core of accurate recommendation. In life, the user's point of interest changes dynamically. It is not only affected by long-term interest, but also has an inseparable relationship with the recent interaction behavior. Traditional recommendation systems make static recommendations based on user interaction history, which cannot capture users' dynamic preferences. The conversational recommendation system aims to describe the user's dynamic interests. It predicts the items that may be interacted with in the future based on the user's click sequence, but it still faces the following problems: [0003] 1. Only use a single type of user behavior data for training, ignoring that the user's future behavior is ...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
IPC IPC(8): G06F16/9535G06F16/9538G06Q30/02
CPCG06Q30/0255G06Q30/0277G06Q30/0631G06F16/9536G06F16/9538
Inventor 王英龙张洪彪舒明雷陈达刘丽孔祥龙
Owner 山东省人工智能研究院
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic, Popular Technical Reports.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com