SciDef: A Research Tool for Automatically Extracting Definitions from Academic Literature Using Large Language Models

SciDef is an automated tool developed by the Media Bias Group based on large language models (LLMs), specifically designed to extract term definitions from academic literature and help researchers quickly understand professional concepts. The project includes a GitHub repository and an academic paper of the same name, aiming to address the issues of time-consuming search for term definitions in academic literature and the difficulty of general dictionaries covering contextualized definitions.

In academic research, professional term definitions are the foundation of literature reading. However, the explosion in the number of academic publications leads to information overload, and a single paper often contains dozens of unfamiliar terms. Traditional manual search for definitions is time-consuming and prone to omissions, and general dictionaries struggle to cover specific, contextualized definitions in literature—this prompted the birth of the SciDef project.

Technical Challenges: Diverse definition forms (formal, operational, exemplary, etc.), term ambiguity (same term has different meanings across disciplines), complex sentence structures in academic texts.

Advantages of LLMs: Possess context understanding capabilities to identify semantic connections between definitions and terms; strong cross-domain generalization ability without needing separate training for each field; can handle complex sentence structures and recognize implicit or scattered definitions.

System Architecture:

1. Document preprocessing: PDF parsing, section-based processing, citation differentiation;
2. Candidate definition identification: Term detection, definition pattern recognition, confidence scoring;
3. Definition extraction and structuring: Boundary determination, relation extraction, machine-readable format output.

Technical Implementation: May adopt prompt engineering, fine-tuning models, and multi-model integration strategies; evaluation metrics include exact matching, semantic equivalence, coverage, precision, and recall.

Application Scenarios:

• Literature review assistance: Quickly extract key term definitions and build knowledge graphs;
• Cross-disciplinary research: Help understand terms from other fields and reduce barriers;
• Academic writing assistance: Check the accuracy of term usage;
• Knowledge base construction: Used for domain-specific knowledge bases or dictionaries.

Connection to Media Bias Research: The Media Bias Group's research requires accurate term definitions (e.g., "bias", "framing"), and SciDef can help organize and standardize the use of key terms.

Current Limitations:

• Domain specificity: Highly specialized fields require additional adaptation;
• Language limitations: Mainly supports English;
• Complex definitions: Extraction of scattered or inference-required definitions is difficult.

Future Directions:

1. Expand multilingual support;
2. Develop domain adaptation layers;
3. Combine manual verification to improve quality;
4. Integrate existing knowledge graphs.

Academic Contributions:

• Clearly define the definition extraction task and provide research benchmarks;
• May construct annotated datasets to promote empirical research;
• Explore the application of LLMs in definition extraction, providing references for subsequent research.

Outlook: SciDef is expected to reduce the information processing burden on researchers and promote knowledge dissemination. As LLM capabilities improve, such tools may become standard for researchers, providing valuable cases for fields like academic information processing.