Deep Fusion of Heterogeneous Features and Semantic Mixer for Graph Neural Network Model

Read full text Summarise with VIRAT
Authors: Jin Zhang
Cited by: 0
References: 19

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

With the wide application of graph neural networks in the field of graph-structured data learning, how to effectively handle the graph data that is ubiquitous in the real world and contains rich heterogeneous features and complex semantics has become a core challenge in current research. Most existing methods focus on processing homogeneous features or single-type node relationships, failing to fully explore and integrate the intrinsic heterogeneous features of nodes (such as text, images, categorical attributes, etc.) and their diverse semantics in different graph contexts. To address this issue, this paper proposes a novel graph neural network model that deeply fuses heterogeneous features and semantic mixers. The model first designs a hierarchical heterogeneous feature alignment and fusion module to map different types of features into a unified semantic space and conduct deep interaction. On this basis, we introduce a multi-perspective semantic mixer, which captures the representations of nodes in different semantic perspectives (such as structural roles, community attributes, global states) through parallel independent semantic perspectives and adaptive perspective fusion mechanisms, and adaptively fuses them. Experimental results on multiple public benchmark datasets show that this model significantly outperforms existing mainstream baseline models in node classification, link prediction, and graph classification tasks, verifying its effectiveness and superiority in handling complex heterogeneous graph data.

Fields of Study & Subjects

Advanced Graph Neural Networks Graph Theory and Algorithms Complex Network Analysis Techniques

Publication details

Journal / source
IEEE Access
Volume
14
Pages
45437–45448
Publication date
1 January 2026
Language
EN