🧠 FinBERT Tone Classification on Earnings Call Transcripts

This repository implements a binary classification model using FinBERT to analyze the tone of earnings call transcripts and predict market sentiment (0 = negative/neutral, 1 = positive).

---

🔍 Overview

• Task: Binary sentiment classification on earnings call transcripts
• Model: ProsusAI/finbert from Hugging Face
• Loss: Weighted BCEWithLogitsLoss to address class imbalance
• Data: Custom cleaned JSONL from earnings calls; optional weak labels

---

---

⚙️ Setup & Requirements

pip install requirements.txt
pip install jsonlines

🏃‍♂️ How to Run

python generate_gold_labels.py
python FinBERTToneEmbeddingClassifier.py

✅Accuracy

📚 Citation

We used ProsusAI/FinBERT and custom cleaned transcripts from the lamini earnings-calls-qa dataset (transcripts only).