kenchanlol/gsoc2024_proposal

Knowledge Graph-Augmented RAG Systems for Improved Frame Blending Generation

Project Overview

This project aims to enhance frame blending generation using knowledge graph-augmented RAG systems. The main goals include:

1. Constructing a comprehensive FrameNet knowledge graph to capture complex relationships such as inter-element relationships and frame hierarchies.

2. Developing a frame blending system with KG-enhanced RAG, improving the system's explainability and controllability, and reducing hallucinations.

3. Developing a chatbot application for frame blending (conditional), allowing interactive generation or refinement of frame blending outcomes.

Technology Focus

• FrameNet: A semantic database essential for mapping meanings in the knowledge graph.
• Large Language Models (LLMs): Key to processing FrameNet data and constructing knowledge graphs.

Plan and Timeline

• Start with enhancing understanding of KG-enhanced RAG and FrameNet.
• Develop a KG Constructor to convert XML data into a Knowledge Graph.
• Build the LLM Frame Blending system using a self-collaboration model approach.
• Implement a Gradio application for visualization and interaction.
• Conclude with testing, documentation, and reporting.

References

• REACT: Synergizing reasoning and acting in language models.
• FrameNet Lexical Semantic Structures for Knowledge Graph Extraction.
• AgentCoder: Multi-Agent-based Code Generation with Iterative Testing and Optimisation.