Deep graph embedding learning based on multi-variational graph autoencoders for POI recommendation

Recently, point-of-interest (POI) recommendation has become a popular research hotspot in heterogeneous location-based social network (LBSN). One major recurring challenge in POI recommendation is that most existing works fail to learn well graph embeddings for user preferences, lacking the capabili...

Full description

Saved in:
Bibliographic Details
Published in:Data mining and knowledge discovery Vol. 39; no. 4; p. 33
Main Authors: Gong, Weihua, Shen, Genhang, Zhao, Anlun, Yang, Lianghuai, Cheng, Zhen
Format: Journal Article
Language:English
Published: New York Springer Nature B.V 01.07.2025
Subjects:
Decoders
Embedding
Graph representations
Graph theory
Learning
Performance evaluation
Social networks
ISSN:1384-5810, 1573-756X
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Recently, point-of-interest (POI) recommendation has become a popular research hotspot in heterogeneous location-based social network (LBSN). One major recurring challenge in POI recommendation is that most existing works fail to learn well graph embeddings for user preferences, lacking the capability of fusing multi-typed nodes and their interaction relations, e.g., users’ check-in relations to POIs, following relations to online topics, and the social relations. To address this challenge, we propose a new unified heterogeneous graph embedding framework by leveraging multimodal variational graph autoencoders, called MultiVGAE. Specifically, we first employ multiple GCN-based encoders to learn the modality-specific latent embeddings for different entities in heterogeneous subgraphs of LBSN, with consideration of fusing multi-types of relations and multi-modal node features. And then reconstruct the corresponding subgraph structures through multiple decoders from the learned embeddings. Finally, extensive experiments have been conducted on two real-world datasets (e.g., Foursquare-NYC and Yelp2018), and the experimental results demonstrate that our proposed MultiVGAE achieves superior performance compared to the existing state-of-the-art baselines on POI recommendation.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:1384-5810
1573-756X
DOI:10.1007/s10618-025-01106-6