| Property | Value |
|---|---|
| Parameters | 110M |
| Architecture | 12-layer, 768-hidden, 12-heads |
| Training Data | 260,773 10-K SEC filings (1993-2019) |
| Paper | FiNER: Financial Numeric Entity Recognition for XBRL Tagging |
What is sec-bert-base?
SEC-BERT-BASE is a specialized BERT model trained specifically for financial domain natural language processing. It's part of the SEC-BERT family of models developed by AUEB's Natural Language Processing Group, designed to enhance financial text analysis capabilities. The model was pre-trained on a massive dataset of SEC filings, making it particularly adept at understanding financial terminology and contexts.
Implementation Details
The model implements a custom 30k subword vocabulary trained from scratch on financial documents. It follows BERT's base architecture but with domain-specific training on financial texts. The training process involved 1 million steps with 256-sequence batches and a 1e-4 learning rate, utilizing Google Cloud TPU v3-8.
- Custom financial vocabulary of 30k subwords
- Pre-trained on 260,773 10-K filings
- Compatible with both PyTorch and TensorFlow 2
- Trained with masked language modeling objective
Core Capabilities
- Superior performance in financial text prediction tasks
- Enhanced understanding of financial terminology
- Accurate numeric value and context prediction
- Improved financial entity recognition
Frequently Asked Questions
Q: What makes this model unique?
SEC-BERT-BASE stands out due to its specialized training on financial documents from SEC filings, making it particularly effective for financial NLP tasks compared to general-purpose BERT models. The model shows significantly better performance in predicting financial contexts and terminology.
Q: What are the recommended use cases?
The model is ideal for financial text analysis tasks including: financial document parsing, numeric entity recognition, financial sentiment analysis, and automated financial report analysis. It's particularly useful for FinTech applications and financial research requiring deep understanding of SEC documents.