| --- |
| license: cc-by-nc-4.0 |
| task_categories: |
| - question-answering |
| language: |
| - en |
| - zh |
| tags: |
| - RAG |
| - Graph |
| - GraphRAG |
| - novels |
| - Multilingual |
| size_categories: |
| - 1K<n<10K |
| configs: |
| - config_name: AnnoyRAG-CHS-QA |
| data_files: |
| - split: test |
| path: annoyrag_chs_qa.parquet |
| - config_name: AnnoyRAG-CHS-Texts |
| data_files: |
| - split: test |
| path: annoyrag_chs_text_chunks.parquet |
| - config_name: AnnoyRAG-ENG-QA |
| data_files: |
| - split: test |
| path: annoyrag_eng_qa.parquet |
| - config_name: AnnoyRAG-ENG-Texts |
| data_files: |
| - split: test |
| path: annoyrag_eng_text_chunks.parquet |
| --- |
| |
| # AnnoyRAG Dataset |
|
|
| The AnnoyRAG dataset, introduced in [Youtu-GraphRAG: Vertically Unified Agents for Graph Retrieval-Augmented Complex Reasoning](https://www.arxiv.org/abs/2508.19855), employs entity anonymization to isolate LLMs' parametric knowledge. This design enables more precise evaluation of how effectively LLMs integrate retrieved information in RAG systems. |
|
|
| ## Dataset Details |
|
|
| #### Dataset Description |
|
|
| The basic statistical information of the dataset is as follows: |
| | **Question Type** | **Difficulty Level** | **Water Margin** | **Dream of <br>Red Chamber** | **Moby-Dick** | **Middlemarch** | **Total** | |
| | :-----------------------: | :---------------------: | :--------------: | :----------------------: | :-----------: | :-------------: | :----------: | |
| | **Objective** | Simple (≤4 hops) | 29 | 26 | 54 | 54 | 163 | |
| | | Complex (>4 hops) | 24 | 34 | 51 | 22 | 131 | |
| | **Subjective<br>(Anonymous Reversion)** | Simple (≤4 hops) | 65 | 50 | 51 | 116 | 282 | |
| | | Complex (>4 hops) | 146 | 314 | 142 | 219 | 821 | |
| | **Total** | - | **264** | **424** | **298** | **411** | **1397** | |
|
|
| - **Languages:** [English&Chinese] |
|
|
| #### Dataset Sources |
|
|
| All raw data in this study are sourced from the original texts of four classic novels: Water Margin, Dream of the Red Chamber, Moby-Dick, and Middlemarch. The copyrights of all these works have entered the public domain. |
|
|
| ## Dataset Structure |
|
|
| #### 1. annoyrag_chs_qa.parquet & annoyrag_eng_qa.parquet |
|
|
| 1. question: questions designed as input queries to assess RAG capabilities. |
| 2. answer: final answers for reference. |
| 3. relations: potentially relevant knowledge graph relations (reference only). |
| 4. entities: potentially relevant knowledge graph entities (reference only). |
| 5. query_type: The question types include multiple-choice questions and de-anonymization tasks. |
| |
| #### 2. annoyrag_chs_text_chunks.parquet & annoyrag_eng_text_chunks.parquet |
| |
| 1. idx: positional indices for chunk ordering. |
| 2. title: section titles after anonymization processing. |
| 3. chunk: The anonymized corpus has been chunked with a maximum size of 1,000 characters per segment. To prevent semantic truncation, we implemented a 100-character overlap between consecutive chunks. |
| |
| ## License |
| - Dataset: cc-by-nc-4.0. |
| - Generated using DeepSeek-V3-0324. |
| |
| ## Citation |
| ``` |
| @misc{dong2025youtugraphragverticallyunifiedagents, |
| title={Youtu-GraphRAG: Vertically Unified Agents for Graph Retrieval-Augmented Complex Reasoning}, |
| author={Junnan Dong and Siyu An and Yifei Yu and Qian-Wen Zhang and Linhao Luo and Xiao Huang and Yunsheng Wu and Di Yin and Xing Sun}, |
| year={2025}, |
| eprint={2508.19855}, |
| archivePrefix={arXiv}, |
| primaryClass={cs.IR}, |
| url={https://arxiv.org/abs/2508.19855}, |
| } |
| ``` |
| |
| |
