CDF-RAG: Causal Dynamic Feedback for Adaptive Retrieval-Augmented Generation

Elahe Khatibi*, Ziyu Wang*, Amir M. Rahmani
University of California, Irvine
📄 arXiv:2504.12560
* Equal contribution

---

🧠 Abstract

Retrieval-Augmented Generation (RAG) has significantly enhanced large language models (LLMs) in knowledge-intensive tasks by incorporating external knowledge retrieval. However, existing RAG frameworks primarily rely on semantic similarity and correlation-driven retrieval, limiting their ability to distinguish true causal relationships from spurious associations. This results in responses that may be factually grounded but fail to establish cause-and-effect mechanisms, leading to incomplete or misleading insights. To address this issue, we introduce Causal Dynamic Feedback for Adaptive Retrieval-Augmented Generation (CDF-RAG), a framework designed to improve causal consistency, factual accuracy, and explainability in generative reasoning. CDF-RAG iteratively refines queries, retrieves structured causal graphs, and enables multi-hop causal reasoning across interconnected knowledge sources. Additionally, it validates responses against causal pathways, ensuring logically coherent and factually grounded outputs. We evaluate CDF-RAG on four diverse datasets, demonstrating its ability to improve response accuracy and causal correctness over existing RAG-based methods.

---

🏗️ Repository Structure

CDF-RAG/
├── agents/               # Core modules (retrieval, rewriting, RL, hallucination detection)
├── data/                 # Dataset processing scripts and Jupyter notebooks
├── eval/                 # Evaluation metrics and scoring scripts
├── finetune/             # Scripts for fine-tuning LLaMA, Mistral, FLAN-T5
├── main/                 # Pipeline runner and ablation experiments
├── figs/                 # Key figures from paper (use .png for rendering)
├── config.py             # Centralized configuration & env variables
├── requirements.txt      # Python dependencies
└── README.md             # You're here

---

🚀 Getting Started

Installation

git clone https://github.com/elakhatibi/CDF-RAG.git
cd CDF-RAG
pip install -r requirements.txt

Environment Setup

Create a .env file or export these:

OPENAI_API_KEY=your-openai-key
PINECONE_API_KEY=your-pinecone-key
NEO4J_URI=bolt://localhost:7687
NEO4J_USER=neo4j
NEO4J_PASSWORD=your-password

---

💡 CDF-RAG System Architecture

The CDF-RAG framework is composed of:

• RL-based query refinement (PPO)
• Dual-path document retrieval using:
  • Semantic similarity from Pinecone
  • Causal graphs from Neo4j
• Knowledge rewriting via GPT
• Hallucination detection & correction using LLM verification

---

📈 Key Results

🔍 Table: Ablation Study

We report the main ablation results of CDF-RAG below. For full quantitative results across 4 datasets and 4 LLMs, refer to the paper.

Ablation Stage	CRC	CCD	SRS	Groundedness	HR	F1
Baseline RAG	0.74	1.50	0.55	0.52	0.18	0.68
+ RL-based Query Refinement	0.80	1.70	0.62	0.59	0.14	0.74
+ Causal Graph	0.84	1.92	0.65	0.63	0.12	0.78
+ Rewriter	0.88	2.00	0.70	0.68	0.08	0.82
+ Hallucination Correction (Ours)	0.89	2.02	0.74	0.71	0.07	0.86

---

🧪 Evaluation Metrics

• CRC: Causal Retrieval Coverage
• CCD: Causal Chain Depth
• SRS: Semantic Refinement Score
• Groundedness: Final answer ↔ retrieved docs
• HR: Hallucination Rate
• F1, Precision, Recall: (if hallucination labels available)

---

🔄 Run the Pipeline

Run a Query

python main/cdf_rag_pipeline.py "Why do patients miss appointments?"

Evaluate Batch of Queries

python main/evaluate_batch_queries.py
python eval/evaluate_metrics.py

Run Ablation Study

python main/Ablation_study.py

---

🏋️‍♂️ Fine-Tuning Instructions

To fine-tune LLaMA, Mistral, or FLAN-T5:

cd finetune/
python finetune_mistral_qlora.py      # or flan/llama variant

All models are instruction-tuned using the same format (instruction/input/output) on causal QA refinement data.

---

📎 Citation

@misc{khatibi2025cdfrag,
    title={CDF-RAG: Causal Dynamic Feedback for Adaptive Retrieval-Augmented Generation},
    author={Elahe Khatibi and Ziyu Wang and Amir M. Rahmani},
    year={2025},
    eprint={2504.12560},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}

---

📬 Contact

For questions or collaborations, feel free to reach out:

• 💬 Elahe Khatibi — ekhatibi@uci.edu
• 💬 Ziyu Wang — ziyuw31@uci.edu

---

❤️ Acknowledgments

This work was supported by the HealthSciTech group at UC Irvine. We thank the LangGraph and open-source LLM communities for inspiration and tooling.

---