AutoBool-Qwen4b-Reasoning-conceptual

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 the conceptual method for structured query construction. The model follows a systematic 5-step process to identify concepts and build Boolean queries based on domain logic.

Base Model: Qwen/Qwen3-4B
Training Method: GRPO (Group Relative Policy Optimization) with LoRA fine-tuning
Prompt Strategy: Conceptual method (structured 5-step approach)
- Step 1: Identify 2-3 key concepts from the topic (e.g., Population, Intervention, Outcome)
- Step 2: For each concept, list related terms (synonyms, variants, relevant MeSH terms)
- Step 3: Create a Boolean block per concept (combine terms using OR)
- Step 4: Use wildcards (*) to capture word variants
- Step 5: Combine all Boolean blocks using AND
- Output format: <think>[Step-by-step conceptual analysis]</think><answer>[Boolean query]</answer>
Domain: Biomedical literature search (PubMed)
Task: Boolean query generation for high-recall retrieval

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
Reasoning Approach: Structured conceptual decomposition

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
Applications requiring systematic concept identification

How to Use

from transformers import AutoTokenizer, AutoModelForCausalLM
import re

model_name = "ielabgroup/Autobool-Qwen4b-Reasoning-conceptual"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)

# Define your systematic review topic
topic = "Ultrasonography for diagnosis of alcoholic cirrhosis in people with alcoholic liver disease"

# Construct the prompt with system and user messages
messages = [
    {"role": "system", "content": "You are an expert systematic review information specialist.
Formulate a systematic review Boolean query using step-by-step reasoning inside <think> </think>, and output the final query inside <answer> </answer>."},
    {"role": "user", "content": f'You are given a systematic review topic titled: "{topic}".
Construct a Boolean query using the **conceptual method**, based on domain logic and structured thinking.

**Step 1**: Identify 2–3 key concepts from the topic (e.g., Population, Intervention, Outcome).

**Step 2**: For each concept:
- List related terms: synonyms, variants, relevant MeSH terms.
- Prioritise specific, high-precision terms.

**Step 3**: Create a Boolean block per concept:
- Combine terms using OR
- Use 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)
- Tag terms individually when needed (e.g., covid-19[ti] vaccine[ti] children[ti])
- Field tags limit search scope and disable ATM

**Step 4**: 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]).

**Step 5**: Combine all Boolean blocks using AND:
((Concept1_term1[tiab] OR Concept1_term2[tiab] OR Concept1_termX[mh]) AND (Concept2_...))

**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 your full reasoning inside <think>...</think>
Output only the final Boolean query inside <answer>...</answer>
Do not include any content outside these tags.
Do not include date limits.'}
]

# 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=4096)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)

# Extract reasoning and query
reasoning_match = re.search(r'<think>(.*?)</think>', response, re.DOTALL)
query_match = re.search(r'<answer>(.*?)</answer>', response, re.DOTALL)

if reasoning_match and query_match:
    reasoning = reasoning_match.group(1).strip()
    query = query_match.group(1).strip()
    print("Step-by-step conceptual analysis:", reasoning)
    print("
Query:", query)

Advantages

Systematic concept-based approach (Population, Intervention, Outcome framework)
Explicit decomposition of complex topics into manageable concepts
Structured 5-step process ensures comprehensive term coverage
Well-suited for systematic reviews with clear conceptual structure

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:

@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={2026}
}