This project addresses the growing problem of AI-generated academic papers containing fabricated references. It provides a tool for researchers and academics to detect potentially hallucinated citations within PDF documents, acting as a crucial line of defense against AI misuse in scholarly work. The primary benefit is enhancing the integrity of academic research by identifying and flagging unreliable references.

How It Works

The tool employs a multi-stage process: it first extracts references from academic PDFs using MuPDF. Subsequently, it validates these extracted references by querying a comprehensive suite of academic databases and resolvers, including CrossRef, arXiv, DBLP, Semantic Scholar, and others, in parallel. For references not found in any database, an optional SearxNG web search fallback can be used for title verification. This approach leverages AI to combat AI-generated content, offering both online API access and robust offline database querying for performance and to circumvent rate limits.

Quick Start & Requirements

• Installation:
  • Rust TUI (recommended): curl -sSf https://hallucinator.science/install.sh | sh
  • Rust CLI: curl -sSf https://hallucinator.science/install-cli.sh | sh
  • Python Bindings: pip install hallucinator (Python 3.12 wheels available for Linux, macOS, Windows).
• Prerequisites:
  • Optional API Keys: OpenAlex, Semantic Scholar, GovInfo for enhanced coverage and reduced rate limiting.
  • Optional Offline Databases: DBLP, arXiv, IACR ePrint, ACL Anthology, OpenAlex require significant disk space and download/build time but offer faster, rate-limit-free querying.
  • Optional SearxNG instance for web search fallback.
• Links:
  • OpenAlex API settings: https://openalex.org/settings/api
  • Semantic Scholar API: https://www.semanticscholar.org/product/api
  • GovInfo API signup: https://api.data.gov/signup/

Highlighted Details

• Supports validation against over 10 academic databases and resolvers, including specialized archives like IACR ePrint.
• Offers extensive offline database support (DBLP, arXiv, etc.) to mitigate API rate limits and improve query speed.
• Automatically checks verified references for retractions using CrossRef metadata and title searches.
• Includes an optional web search fallback via SearxNG for references not found in academic databases.

Maintenance & Community

This project was created by Gianluca Stringhini with assistance from AI models. No specific community channels (e.g., Discord, Slack) or roadmap details are provided in the README.

Licensing & Compatibility

The project is licensed under the GNU Affero General Public License v3.0 (AGPL-3.0). This is a strong copyleft license, meaning that modifications or derivative works, especially when distributed or used in a network service, may require making the source code available under the same terms.

Limitations & Caveats

Database coverage is not exhaustive, and very recent papers may not yet be indexed. The tool may struggle with non-academic content like books or technical reports, and poor PDF formatting can hinder reference extraction. While web search fallback is available, it only verifies titles, not authors, and requires manual confirmation. Some references (e.g., short titles, URLs) are intentionally skipped.