FinBERT is a BERT language model pre-trained on a large corpus of financial communications, designed to advance financial Natural Language Processing (NLP) research and applications. It offers specialized models for sentiment analysis, ESG classification, and forward-looking statement (FLS) classification, outperforming traditional and other deep learning models on these tasks.

How It Works

FinBERT leverages the BERT architecture, pre-trained on 4.9 billion tokens from corporate reports (10-K & 10-Q), earnings call transcripts, and analyst reports. This extensive financial corpus allows FinBERT to capture domain-specific language nuances. The project also provides fine-tuned versions of the model for specific NLP tasks, demonstrating state-of-the-art performance.

Quick Start & Requirements

• Install/Run: Use Huggingface's transformers library.
• Prerequisites: Python, Huggingface transformers, torch, numpy. No specific hardware requirements beyond standard ML inference.
• Demo: FinBERT-demo.ipynb and finetune.ipynb are provided in the repository.
• More Info: FinBERT.AI

Highlighted Details

• Offers four pre-trained FinBERT versions: FinVocab-Uncased (recommended), FinVocab-Cased, BaseVocab-Uncased, and BaseVocab-Cased.
• Fine-tuned models available on Huggingface for sentiment, ESG, and FLS classification.
• Achieves state-of-the-art performance on various financial NLP tasks.
• Pre-trained on a 4.9B token financial corpus including 10-K/10-Q reports, earnings calls, and analyst reports.

Maintenance & Community

• Migrated to Huggingface in July 2021.
• Contact: imyiyang@ust.hk or GitHub issues.

Licensing & Compatibility

• The README does not explicitly state a license. The code examples are compatible with Huggingface's transformers library, which typically uses Apache 2.0.

Limitations & Caveats

The project's license is not explicitly stated in the README, which may pose a risk for commercial use or closed-source integration.