MindMap is a plug-and-play prompting framework designed to enable Large Language Models (LLMs) to perform graph-of-thoughts reasoning. It allows LLMs to comprehend graphical inputs and construct their own mind maps, facilitating evidence-grounded generation. This approach is beneficial for researchers and developers working with LLMs who need to enhance their reasoning capabilities with structured knowledge.

How It Works

MindMap leverages knowledge graph prompting to guide LLMs in building internal mind maps. This method allows the LLM to process and integrate graphical information, supporting a more structured and evidence-based generation process. The framework aims to spark a "graph-of-thoughts" within the LLM, moving beyond linear reasoning.

Quick Start & Requirements

• Installation: Clone the repository and modify MindMap.py to include your Neo4j Sandbox URI, username, and password, as well as your OpenAI API key.
• Prerequisites:
  • Neo4j Sandbox account and credentials.
  • OpenAI API key.
• Setup: Initial setup involves configuring API keys and Neo4j connection details. The README notes that loading the full CMCKG dataset can take approximately two days, but the EMCKG for the chatdoctor5k example is readily available.
• Documentation: ACL'24 Paper (linked via citation).

Highlighted Details

• Enables LLMs to comprehend graphical inputs.
• Facilitates evidence-grounded generation.
• Plug-and-play prompting approach.
• Sparks "graph-of-thoughts" reasoning in LLMs.

Maintenance & Community

The project is associated with authors Yilin Wen, Zifeng Wang, and Jimeng Sun. Further community or maintenance details are not provided in the README.

Licensing & Compatibility

The licensing information is not explicitly stated in the provided README. Compatibility for commercial use or closed-source linking is not specified.

Limitations & Caveats

The README mentions that loading the full CMCKG dataset is time-consuming (approx. two days), though the EMCKG for the example dataset is faster. Specific limitations regarding model compatibility or performance benchmarks are not detailed.