A multi-task heterogeneous graph learning with cross-attention fusion for robust recommendation.

Area of Science:

• Artificial Intelligence
• Computer Science
• Machine Learning

Background:

• Graph neural networks (GNNs) excel at modeling user-item interactions but struggle with data sparsity, cold-start issues, and conflicting learning objectives.
• Existing GNN-based recommender systems often fail to leverage rich semantic information or effectively manage multiple, potentially interfering, learning tasks.

Purpose of the Study:

• To propose a multi-task heterogeneous graph representation framework that overcomes limitations of current GNN-based recommender systems.
• To enhance recommender system performance by addressing data sparsity, cold-start scenarios, and negative interference among heterogeneous learning objectives at both data and model levels.

Main Methods:

• Enriched the MovieLens user-item interaction graph with semantic information from the IMDb knowledge graph, creating a multi-type heterogeneous graph.
• Designed a dual-branch architecture featuring a multi-task Graph Convolutional Network (GCN) for interaction modeling and a multi-task Heterogeneous Graph Transformer (HGT) for semantic representation.
• Integrated representations using a bidirectional cross-attention mechanism and fine-tuned the fused model end-to-end for joint recommendation and classification tasks.

Main Results:

• The proposed framework demonstrated more stable learning and superior performance compared to conventional GNN-based recommender models.
• Effectively addressed challenges including data sparsity, cold-start scenarios, and negative interference among learning objectives.
• Validated the effectiveness of combining heterogeneous graph enrichment, task-oriented multi-task learning, and cross-attention feature fusion.

Conclusions:

• The novel multi-task heterogeneous graph framework significantly improves recommender system performance and learning stability.
• Integrating diverse data sources and employing a dual-branch architecture with cross-attention fusion offers a robust solution for complex recommendation tasks.
• The study highlights the potential of advanced GNN techniques for building more effective and versatile recommender systems.