133 lines
5.4 KiB
Markdown
133 lines
5.4 KiB
Markdown
---
|
|
license: apache-2.0
|
|
base_model: Qwen/Qwen3-4B
|
|
tags:
|
|
- boolean-queries
|
|
- systematic-review
|
|
- information-retrieval
|
|
- pubmed
|
|
- reinforcement-learning
|
|
- grpo
|
|
library_name: transformers
|
|
---
|
|
|
|
# AutoBool-Qwen4b-No-reasoning
|
|
|
|
This model is part of the **AutoBool** framework, a reinforcement learning approach for training large language models to generate high-quality Boolean queries for systematic literature reviews.
|
|
|
|
## Model Description
|
|
|
|
This variant uses **direct generation** without explicit reasoning steps. The model is instructed to output only the final Boolean query inside `<answer></answer>` tags without any explanation or reasoning process.
|
|
|
|
- **Base Model:** Qwen/Qwen3-4B
|
|
- **Training Method:** GRPO (Group Relative Policy Optimization) with LoRA fine-tuning
|
|
- **Prompt Strategy:** Direct generation (no reasoning)
|
|
- System instruction: "Do not include any explanation or reasoning"
|
|
- Output format: `<answer>[Boolean query]</answer>`
|
|
- No intermediate thinking or explanation steps
|
|
- **Domain:** Biomedical literature search (PubMed)
|
|
- **Task:** Boolean query generation for high-recall retrieval
|
|
|
|
## 🚀 Interactive Demo
|
|
|
|
Try out our query generation models directly in your browser! The demo allows you to test our different reasoning strategies (Standard, Conceptual, Objective, and No-Reasoning) in real-time.
|
|
|
|
[](https://huggingface.co/spaces/wshuai190/AutoBool-Demo)
|
|
* **Live Demo:** [AutoBool on Hugging Face Spaces](https://huggingface.co/spaces/wshuai190/AutoBool-Demo)
|
|
|
|
|
|
## Training Details
|
|
|
|
The model was trained using:
|
|
- **Optimization:** GRPO (Group Relative Policy Optimization)
|
|
- **Fine-tuning:** LoRA (Low-Rank Adaptation)
|
|
- **Dataset:** wshuai190/pubmed-pmc-sr-filtered
|
|
- **Reward Function:** Combines syntactic validity, format correctness, and retrieval effectiveness
|
|
|
|
## Intended Use
|
|
|
|
This model is designed for:
|
|
- Generating Boolean queries for systematic literature reviews
|
|
- High-recall biomedical information retrieval
|
|
- Supporting evidence synthesis in healthcare and biomedical research
|
|
|
|
## How to Use
|
|
|
|
```python
|
|
from transformers import AutoTokenizer, AutoModelForCausalLM
|
|
|
|
model_name = "ielabgroup/Autobool-Qwen4b-No-reasoning"
|
|
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
|
model = AutoModelForCausalLM.from_pretrained(model_name)
|
|
|
|
# Define your systematic review topic
|
|
topic = "Thromboelastography (TEG) and rotational thromboelastometry (ROTEM) for trauma-induced coagulopathy"
|
|
|
|
# Construct the prompt with system and user messages
|
|
messages = [
|
|
{"role": "system", "content": "You are an expert systematic review information specialist.
|
|
You are tasked to formulate a systematic review Boolean query in response to a research topic. The final Boolean query must be enclosed within <answer> </answer> tags. Do not include any explanation or reasoning."},
|
|
{"role": "user", "content": f'You are given a systematic review research topic, with the topic title "{topic}".
|
|
Your task is to formulate a highly effective Boolean query in MEDLINE format for PubMed.
|
|
The query should balance **high recall** (capturing all relevant studies) with **reasonable precision** (avoiding irrelevant results):
|
|
- Use both free-text terms and MeSH terms (e.g., chronic pain[tiab], Pain[mh]).
|
|
- **Do not wrap terms or phrases in double quotes**, as this disables automatic term mapping (ATM).
|
|
- Combine synonyms or related terms within a concept using OR.
|
|
- Combine different concepts using AND.
|
|
- Use wildcards (*) to capture word variants (e.g., vaccin* → vaccine, vaccination):
|
|
- Terms must have ≥4 characters before the * (e.g., colo*)
|
|
- Wildcards work with field tags (e.g., breastfeed*[tiab]).
|
|
- Field tags limit the search to specific fields and disable ATM.
|
|
- Do not include date limits.
|
|
- Tag term using term field (e.g., covid-19[ti] vaccine[ti] children[ti]) when needed.
|
|
**Only use the following allowed field tags:**
|
|
Title: [ti], Abstract: [ab], Title/Abstract: [tiab]
|
|
MeSH: [mh], Major MeSH: [majr], Supplementary Concept: [nm]
|
|
Text Words: [tw], All Fields: [all]
|
|
Publication Type: [pt], Language: [la]
|
|
|
|
Output and only output the formulated Boolean query inside <answer></answer> tags. Do not include any explanation or content outside or inside the <answer> tags.'}
|
|
]
|
|
|
|
# Generate the query
|
|
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
|
|
inputs = tokenizer(prompt, return_tensors="pt")
|
|
outputs = model.generate(**inputs, max_length=2048)
|
|
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
|
|
|
# Extract the query from <answer> tags
|
|
import re
|
|
match = re.search(r'<answer>(.*?)</answer>', response, re.DOTALL)
|
|
if match:
|
|
query = match.group(1).strip()
|
|
print(query)
|
|
```
|
|
|
|
## Limitations
|
|
|
|
- Optimized specifically for PubMed Boolean query syntax
|
|
- Performance may vary on non-biomedical domains
|
|
- Requires domain knowledge for effective prompt engineering
|
|
|
|
## Citation
|
|
|
|
If you use this model, please cite:
|
|
|
|
```bibtex
|
|
@inproceedings{autobool2026,
|
|
title={AutoBool: Reinforcement Learning for Boolean Query Generation in Systematic Reviews},
|
|
author={[Shuai Wang, Harrisen Scells, Bevan Koopman, Guido Zuccon]},
|
|
booktitle={Proceedings of the 2026 Conference of the European Chapter of the Association for Computational Linguistics (EACL)},
|
|
year={2025}
|
|
}
|
|
```
|
|
|
|
## More Information
|
|
|
|
- **GitHub Repository:** [https://github.com/ielab/AutoBool](https://github.com/ielab/AutoBool)
|
|
- **Paper:** Accepted at EACL 2026
|
|
|
|
## License
|
|
|
|
Apache 2.0
|