MEHGT-LKG Dataset

Dataset Summary

MEHGT-LKG provides the core multimodal financial data resources used in our framework for stock trend prediction. The current release includes raw financial news data, stock trading data, and northbound capital flow data. These resources support financial event extraction, multimodal financial modeling, and downstream stock trend prediction research.

This repository is intended for downloading, training, and academic study. We will continue to expand the dataset in future releases, for example by covering more stocks and enriching the scale of multimodal financial data.

Dataset Structure

The repository currently includes the following components:

• news_data: raw financial news data used for financial event extraction and financial text analysis;
• Trading_data: stock trading data and related numerical market information;
• HK-SH&SZ: northbound capital flow data associated with cross-market trading activity.

These data resources are designed to support the multimodal setting of MEHGT-LKG, where financial news, trading signals, and capital-flow information are jointly modeled for stock trend prediction.

Intended Use

This dataset is intended for research on:

• stock trend prediction,
• financial event extraction,
• multimodal financial learning,
• heterogeneous graph-based financial modeling,
• academic training and reproducible experiments.

Support

• The dataset is mainly constructed for financial research scenarios and currently covers a limited set of stocks and data sources.
• Future versions will continue to expand the stock universe and enrich the multimodal financial data.

Citation

If you use this dataset, please cite our work:

@misc{ zhou2026mehgtlkg, title={{MEHGT}-{LKG}: Multimodal Edge-enhanced Heterogeneous Graph Transformer with {LLM}-driven Knowledge Graph for Stock Trend Prediction}, author={Quanshi Zhou and Wei Chen and Ni An and Manrui Jiang}, year={2026}, url={https://openreview.net/forum?id=N5ggpxl8Os} }