Scholarlink

<p>Heterogeneous graph representation learning has attracted considerable interest for its effectiveness in capturing the structural information of nodes within complex and diverse networks. However, existing heterogeneous graph representation learning models primarily focus on structural embeddings, overlooking the importance of node attributes and temporal information in understanding complex interaction networks, particularly in e-commerce heterogeneous networks. To address this issue, this paper proposes an <b>A</b>ttribute-enhanced <b>M</b>ulti-<b>I</b>nteraction <b>R</b>epresentation <b>L</b>earning method in <b>e</b>-commerce heterogeneous networks, namely AMIRLe. Specifically, AMIRLe first processes high-dimensional data and identifies hidden patterns via tensor decomposition techniques to construct multi-interaction heterogeneous networks. Then, a novel time-aware attribute embedding method is proposed, which employs an attention mechanism to enhance key node aggregation after integrating interaction time information as auxiliary attributes with node attributes. Finally, a graph neural network (GNN) method is designed to integrate meta-path-based structure embeddings with attribute embeddings to produce the final embeddings. Link prediction experiments on three real-world e-commerce datasets demonstrate that performance of AMIRLe superior to state-of-the-art methods.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

AMIRLe: attribute-enhanced multi-interaction representation learning for e-commerce heterogeneous networks

  • 期刊论文

  • 发表日期：

    2026年2月28日

查看全文
  • Lianhong Ding,
  • Mengxiao Li,
  • Yi Wang,
  • Peng Shi,
  • Fangfeng Zhang

摘要

Heterogeneous graph representation learning has attracted considerable interest for its effectiveness in capturing the structural information of nodes within complex and diverse networks. However, existing heterogeneous graph representation learning models primarily focus on structural embeddings, overlooking the importance of node attributes and temporal information in understanding complex interaction networks, particularly in e-commerce heterogeneous networks. To address this issue, this paper proposes an Attribute-enhanced Multi-Interaction Representation Learning method in e-commerce heterogeneous networks, namely AMIRLe. Specifically, AMIRLe first processes high-dimensional data and identifies hidden patterns via tensor decomposition techniques to construct multi-interaction heterogeneous networks. Then, a novel time-aware attribute embedding method is proposed, which employs an attention mechanism to enhance key node aggregation after integrating interaction time information as auxiliary attributes with node attributes. Finally, a graph neural network (GNN) method is designed to integrate meta-path-based structure embeddings with attribute embeddings to produce the final embeddings. Link prediction experiments on three real-world e-commerce datasets demonstrate that performance of AMIRLe superior to state-of-the-art methods.