Knowledge Graph and Deep Learning-based Text-to-GQL Model for Intelligent Medical Consultation Chatbot

Area of Science:

• Natural Language Processing
• Graph Databases
• Human-Robot Interaction

Background:

• Existing semantic parsing methods primarily focus on Text-to-SQL, leaving a gap for graph databases.
• Efficient human-machine communication requires translating natural language into structured queries.
• Medical Human-Robot Interactions (HRI) can benefit from improved natural language understanding for graph data.

Purpose of the Study:

• To propose the Text-to-GQL (Text2GQL) task for semantic parsing in graph databases.
• To develop a pipeline solution for the Text2GQL task to bridge the gap in existing research.
• To enhance medical HRI by enabling more direct human-machine communication with graph data.

Main Methods:

• A novel pipeline integrating a Language Model, a Pre-trained Adapter plug-in, and a Pointer Network.
• The Adapter is pre-trained on linking GQL schemas with utterances for knowledge introduction.
• The model copies tokens from utterances and generates GQL statements, with an adjustment mechanism for improved output.

Main Results:

• The proposed Text2GQL model demonstrates competitiveness on converted Text-to-SQL datasets (Spider, ATIS, GeoQuery, 39.net).
• The method proves practical for applications in medical scenarios.
• The pipeline effectively translates natural language questions into executable GQL statements.

Conclusions:

• The developed Text2GQL pipeline offers a viable solution for semantic parsing in graph databases.
• The approach enhances the directness and efficiency of human-machine communication.
• This work lays the foundation for advanced applications in fields like medical HRI.