Stock Recognizer 📈

Introduction

A hybrid Python library for extracting stock tickers and company names from messy social media text (Reddit, Twitter, etc.) — fast regex + market-data matching, with a GLiNER2 AI model layered on by default for higher recall. Pass use_ai=False for a lightweight, dependency-light instance when you only need the regex path.

The goal is to answer "what stocks is this post talking about?" — recognize() returns a deduplicated set of tickers per text. A symbol mentioned once or twenty times yields the same result; the focus is coverage of which tickers appear, not counting every occurrence. Accuracy is measured the same way (set-based, deduplicated per document), so a ticker only needs to be caught once to count as found.

Table of Contents 🗂

• Key Features
• Installation
• Usage
• Model Training
• Citation
• Contributing
• License

Key Features 🔑

• Hybrid Engine: Combines Regex, financedatabase (Market Data), and GLiNER2 (AI).
• Context Aware: Distinguishes between "DD" (Due Diligence) and "DD" (DuPont).
• Yelling Protection: Smart filters for posts written in ALL CAPS.
• Auto-Mapping: Automatically converts "Apple" or "TSMC" to AAPL and TSM.

Installation ⚙️

Installation can be done via pip:

pip install stock-recognizer

Usage ⌨️

from stock_recognizer import StockRecognizer

# Default: use_ai=True, most accurate.
# First run downloads the fine-tuned adapter from the HF Hub (cached after that)
recognizer = StockRecognizer()

text = "$PLAB DD: easy to understand TSMC supplier"
tickers_ai = recognizer.recognize_ai(text)
print(tickers_ai) # ['PLAB', 'TSM']

# Opt out of AI for a lighter, faster, regex + market-data-only path
fast_recognizer = StockRecognizer(use_ai=False)
tickers = fast_recognizer.recognize(text)
print(tickers) # ['PLAB'] (TSMC needs AI mapping)

Model Training 🧠

The GLiNER2 adapter behind recognize_ai() — scraping, labeling policy, training, and benchmarking — lives in the sibling stock-recognizer-model repo. StockRecognizer() (use_ai=True by default) fetches the trained adapter automatically from StephanAkkerman/stock-recognizer-model on the Hugging Face Hub (pass adapter_path= to use a local adapter instead, or adapter_revision= to pin a different published version).

Citation ✍️

If you use this project in your research, please cite as follows:

@misc{project_name,
  author  = {Stephan Akkerman},
  title   = {Stock Recognizer},
  year    = {2026},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/StephanAkkerman/stock-recognizer}}
}

Contributing 🛠

Contributions are welcome! If you have a feature request, bug report, or proposal for code refactoring, please feel free to open an issue on GitHub. We appreciate your help in improving this project.

License 📜

This project is licensed under the MIT License. See the LICENSE file for details.