Messages are Never Propagated Alone: Collaborative Hypergraph Neural Network for Time-Series Forecasting.

Area of Science:

• Data Science
• Machine Learning
• Artificial Intelligence

Background:

• Correlated time-series forecasting is crucial for applications like stock prediction and traffic analysis.
• Conventional graph structures struggle to represent complex, non-pairwise relationships in data.
• Existing methods lack the ability to capture intricate interactions effectively.

Purpose of the Study:

• To introduce a novel hypergraph neural network model, CHNN, for enhanced correlated time-series forecasting.
• To address the limitations of conventional graph structures in representing complex data interactions.
• To improve the accuracy and robustness of time-series forecasting in practical scenarios.

Main Methods:

• Utilizing dynamic hypergraphs to model intricate, non-pairwise relationships.
• Developing the CHNN model incorporating semantic and topological similarities.
• Implementing an interaction model and hypergraph diffusion for correlation scores.
• Integrating short-term and long-term temporal modules with attention and recurrent networks.

Main Results:

• CHNN effectively captures complex spatio-temporal dependencies in correlated time-series data.
• The model leverages semantic and topological similarities for comprehensive correlation scores.
• Experimental results on four real-world datasets show significant performance improvements over benchmarks.
• CHNN demonstrates superior accuracy in forecasting tasks.

Conclusions:

• Dynamic hypergraphs provide a powerful framework for modeling complex interactions in time-series data.
• The proposed CHNN model offers a significant advancement in correlated time-series forecasting.
• CHNN's ability to integrate semantic, topological, and temporal information leads to superior predictive performance.