Multi-Source Financial & General News
- Type
- corpus
- Venue
- Brianferrell787 / Hugging Face
- Year
- 2026
- Source
- huggingface
- Access
- restricted
- Language
- English
- Added
- 2026-07-17T20:18:03.694896+00:00
- Verified
- 2026-07-17T20:18:03.694896+00:00
Summary
Multi-Source Financial & General News is a unified corpus aggregating 24 public news datasets into one consistent, ready-to-use layer totaling 57.1 million rows spanning 1990–2025. All subsets share a minimal schema (date, text, extra_fields) and are shipped as streamable Parquet shards with a trading-date policy to prevent look-ahead bias. Sources include Benzinga, Bloomberg/Reuters, NYT, CNBC, Yahoo Finance, Reddit WorldNews, DJIA stock headlines, and others. The dataset is designed for stock trading reinforcement learning, financial NLP, event studies, and language modeling, addressing the lack of standardized datasets in RL+NLP for stock trading.
Keywords
finance news stock-trading reinforcement-learning nlp time-series markets trading
Topics
Finance / NLP / News
Research notes
- Maintainer: ferrellbj@vcu.edu. Associated with arxiv: 2112.02095 and DOI 10.57967/hf/6432. Requires accepting non-commercial research use terms. Trading date policy anchors day-level news to next NYSE trading date to prevent look-ahead bias.