--- language: - en license: apache-2.0 pretty_name: rare disease corpus tags: - rare disease corpus - rare-disease corpus - rare-disease database dataset_info: features: - name: id dtype: string - name: title dtype: string - name: content dtype: string - name: contents dtype: string - name: nordid dtype: int64 - name: rare-disease dtype: string splits: - name: train num_bytes: 34808885 num_examples: 9268 download_size: 17060625 dataset_size: 34808885 configs: - config_name: default data_files: - split: train path: data/train-* --- # Dataset Card for ReCOP ## What's for? The data for ReCOP is sourced from the [National Organization for Rare Disorders (NORD) database](https://rarediseases.org/), which compiles reports on rare diseases. __NORD is committed to the identification, treatment, and cure of rare diseases through education, advocacy, research, and service programs.__ The primary objective of developing ReCOP using the NORD database is to provide comprehensive expertise on rare diseases for LLMs. This expertise can be leveraged to enhance the diagnostic capabilities of LLMs through retrieval-augmented generation. ## Corpus Overview ReCOP divides each rare disease report into chunks: __overview, symptoms, causes, effects, related disorders, diagnosis__, and __standard therapies__. Each property of the disease corresponds to a specific chunk in ReCOP. In this manner, ReCOP generates 9268 chunks based on the reports of 1324 rare diseases for the NORD database, with each report producing seven chunks corresponding to the properties of a rare disease. ## Using ReCOP for Retrieval Augmentation Generations Simply follow our benchmark repository [**ReDis-QA-Bench**](https://github.com/guanchuwang/redis-bench) to run the retrieval augmentation generations on the [ReDis-QA](https://huggingface.co/datasets/guan-wang/ReDis-QA) dataset: ```bash git clone https://github.com/guanchuwang/redis-bench.git cd redis-bench bash rag-bench/scripts/run_exp.sh ``` ## Benchmark Results of Retrieval Augmentation Generations Benchmark results of retrieval augmentation generations based on ReCOP, where the LLMs take [Llama-2-7B-chat](https://huggingface.co/meta-llama/Llama-2-7b-chat), [Mistral-7B-instruct-v0.2](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2), [Phi-3-7B-instruct](https://huggingface.co/microsoft/Phi-3-small-8k-instruct), [Gemmma-1.1-7B-it](https://huggingface.co/google/gemma-1.1-7b-it), and [Qwen-2-7B-Instruct](https://huggingface.co/Qwen/Qwen2-7B-Instruct). ## Citation Information If you find this corpus useful to your project, we appreciate you citing this work: ```` @article{wang2024assessing, title={Assessing and Enhancing Large Language Models in Rare Disease Question-answering}, author={Wang, Guanchu and Ran, Junhao and Tang, Ruixiang and Chang, Chia-Yuan and Chuang, Yu-Neng and Liu, Zirui and Braverman, Vladimir and Liu, Zhandong and Hu, Xia}, journal={arXiv preprint arXiv:2408.08422}, year={2024} } ````