---
language: en
license: apache-2.0
tags:
- finance
- esg
- sentiment-analysis
- bert
metrics:
- f1
- accuracy
---

# finbert_esg_sentiment_classifier

## Overview
This model is a specialized BERT-based classifier fine-tuned for Environmental, Social, and Governance (ESG) sentiment analysis in financial reports. It categorizes text into specific ESG pillars or identifies neutral financial statements.



## Model Architecture
The model utilizes a **BERT-Base-Uncased** backbone with a sequence classification head. 
- **Encoder**: 12-layer Transformer.
- **Hidden Dimensions**: 768.
- **Head**: Linear layer followed by Softmax for 4-class categorization.
- **Optimization**: Trained using the Cross-Entropy loss function:
$$\mathcal{L} = -\sum_{c=1}^{M} y_{o,c} \ln(p_{o,c})$$

## Intended Use
- **Investment Research**: Automating the extraction of ESG signals from 10-K filings and earnings transcripts.
- **Compliance**: Monitoring corporate communications for ESG-related disclosures.
- **Sustainable Finance**: Providing data for ESG scoring algorithms.

## Limitations
- **Context Window**: Restricted to 512 tokens. Long documents must be processed in chunks.
- **Language**: Optimized for English financial terminology; performance on other languages or casual text is not guaranteed.
- **Factuality**: Classification is based on linguistic patterns, not external fact-checking of the corporate claims.