AgenticSearchQueryset/ASQ

---
license: mit
task_categories:
- question-answering
- text-generation
language:
- en
tags:
- agent
- ir
- retrieval
- search
- rag
- text-generation
pretty_name: '🤖 ASQ: Agentic Search Queryset'
size_categories:
- 100M<n<1B
viewer: false
---

<div align="center">
<h1>🤖 ASQ: Agentic Search Queryset </h1>
<p><strong> A dataset capturing RAG agents' search behaviours.</strong></p>
</div>

<div align="center">

[](https://arxiv.org/pdf/2602.17518)
[](https://github.com/fpezzuti/ASQ)
[](https://opensource.org/licenses/MIT)

</div>

---

## 📖 Dataset Description
**ASQ** (Agentic Search Queryset) is a dataset designed to capture the **search behaviors** of the **RAG agents**.
It collects intermediate **synthetic queries**, **retrieved documents**, and **thoughts** (reasoning descriptions) produced or consumed by agents.

### 📊 Dataset Statistics
- 615k traces (0.12% incomplete)
- 614k answers
- 680k synthetic queries
- 680k retrieved ranked lists
- 3 diverse agent settings
- 2 diverse retrieval settings

---

## 🏗️ Data Construction

Details about **dataset construction** are available on the **paper**: [arXiv preprint](https://arxiv.org/pdf/2602.17518).

See our **GitHub repository** to **reproduce** the construction of ASQ, or to **extend** it:[github.com/fpezzuti/ASQ](https://github.com/fpezzuti/ASQ).

---

## 📂 Dataset Organisation
The ASQ dataset is organised hierarchically under the `traces/` directory:

```markdown
traces/
└── dataset/
└── retriever_config/
└── agent_family/
└── model/
├── answers.tsv
├── iter_queries.tsv
├── retrieved_docs.tsv
└── thoughts.tsv
```
Placeholders:
- dataset: the base dataset from which traces were collected (e.g., HotpotQA-test).
- retriever_config: retrieval pipeline's configuration (e.g., BM25_k100_electra_k3).
- agent_family: type of agent generating the traces (e.g., Autorefine).
- model: generator variant (e.g., Qwen-7B).

---

## 📦 Artifacts
Each collection of traces comprises four TSV artifact files, with rows associated with a qid of the organic query:
| Artifact | Description |
|-------|-------------|
| `answers.tsv` | Answers generated by the agent for each query. |
| `iter_queries.tsv` | Synthetic queries generated by the agent per query and iteration. |
| `retrieved_docs.tsv` | Ranked list of documents retrieved per query and iteration. |
| `thoughts.tsv` | CoT reasoning "thoughts" of the agent per query and iteration.|

### 🏁 Answers
Columns:
- `qid` (string): qid of the organic query.
- `answer` (string): answer generated by the agent.

### ✍️ Synthetic Queries
Columns:
- `qid` (string): qid of the organic query.
- `iteration` (integer): iteration number within agent's inference loop.
- `llm_query` (string): synthetic query generated at that iteration.

### 🔍 Retrieved Documents
Columns:
- `qid` (string): qid of the organic query.
- `iteration` (integer): iteration number within the agent's inference loop.
- `docid` (integer): identifier of the retrieved document.
- `rank` (integer): rank of the document in the ranked list retrieved during that iteration.

### 💭 Thoughts
Columns:
- `qid` (string): qid of the organic query.
- `iteration` (integer): iteration number within the agent's inference loop.
- `thought` (string): chain-of-thought thought produced by the agent at that iteration.

---

## 🚀 Usage
Please see our **GitHub repository**:[github.com/fpezzuti/ASQ](https://github.com/fpezzuti/ASQ).

---

## ⚖️ License
The ASQ dataset is released under the [*MIT License*](https://opensource.org/license/mit). Individual source datasets may have their own licenses.

---

## 🛠️ Ethics Statement
ASQ is derived from publicly available datasets and is intended solely for research on agentic search behaviour.
The authors do not endorse or assume responsibility for the content or any biases present in the traces.
The contents of these traces should not be interpreted as representing the views of the researchers or their institutions.
Users are advised to apply safety and content filters when using them.

---
## 🔗 Citation

If you find our work useful, please cite it as follows:

```bibtex
@misc{fpezzuti2026asq,
title={A Picture of Agentic Search},
author={Pezzuti, Francesca and Frieder, Ophir and Silvestri, Fabrizio and MacAvaney, Sean and Tonellotto, Nicola},
year={2026},
eprint={2602.17518},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/pdf/2602.17518},
}
```