Think Wider, Detect Sharper

Reinforced Reference Coverage for Document-Level Self-Contradiction Detection

This repository contains the official implementation of the paper: Think Wider, Detect Sharper: Reinforced Reference Coverage for Document-Level Self-Contradiction Detection

---

📦 Install Environment

pip install -r requirements.txt

---

🚀 Quick Start

1. 🏗️ Data Construction

Training data is constructed with StorySumm, REPLIQA, and CoT Distillation using DeepSeek R1.

bash script/data_constructor.sh

---

2. 🧠 Train the Model (SFT + RL)

Step 1: Supervised Fine-Tuning (SFT)

bash script/sft.sh

Step 2: Reinforcement Learning Fine-Tuning (RL)

bash script/rl.sh

---

3. 📊 Evaluate the Model

Evaluate (Base, CoT, SFT, RL) on the ContraDoc dataset.

bash script/test.sh