Multi-Table Question Answering with Dynamic Schema Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing question answering systems, particularly those using large language models (LLMs) for multi-table question answering, require significant computational resources, are not scalable, and are limited by fixed schemas, making them inefficient and costly to adapt to diverse databases and domains.
Innovation Solution
A multi-table question answering system that decomposes complex questions into sub-questions using coreference resolution, dependency parsing, and chunking, generates code scripts with LLMs, and executes these scripts to retrieve structured data from multiple tables, dynamically adapting to schema descriptions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If fixed schemas are used to dictate the structure of data tables, then the system structure is simplified and easier to implement, but the adaptability of the system to different databases and domains is limited
Solution Approach 1:
The patent applies dynamics by replacing fixed schemas with dynamic schema inference. The system automatically infers database schemas from natural language descriptions and adapts to different table structures without manual configuration. This allows the same system to work across diverse databases and domains while maintaining operational simplicity through automation.
Solution Approach 2:
The system performs self-service by automatically inferring schemas and generating queries without requiring manual schema configuration or user intervention. The model self-adapts to different database structures by learning from the provided natural language descriptions, eliminating the need for complex manual setup while maintaining high adaptability.
2Reliability
If large language models are fine-tuned for specific tables or databases, then the question answering accuracy is improved, but the cost and scalability are increased
Solution Approach 1:
The patent applies universality by training a single LLM on diverse multi-table question answering tasks across different domains. The model learns general patterns for inferring schemas and generating queries that work across various database structures. This universal approach maintains high accuracy without requiring separate fine-tuning for each domain, thereby improving scalability and reducing costs.
Solution Approach 2:
The system changes parameters dynamically by adjusting the model's inference behavior based on the specific database schema and query requirements. Rather than fine-tuning the entire model for each domain, the system adapts its parameters and reasoning process to match the specific table structures and question types, maintaining accuracy while avoiding the computational cost of full fine-tuning.
3Loss of information
If complex questions are answered by retrieving data from multiple tables, then the completeness of the answer is improved, but the computational resources and time required increase
Solution Approach 1:
The patent applies segmentation by breaking down complex multi-table queries into smaller, manageable sub-queries. The system divides the information retrieval task across multiple tables and uses intermediate representations to connect different data sources. This segmentation allows the model to retrieve necessary information systematically while reducing the computational burden by processing queries in manageable chunks rather than attempting to process the entire complex query at once.
Data Source
AI summary
A method and system for multi-table question answering is provided herein. The method includes receiving a complex question from a user. The method may further include decomposing the complex question into one or more sub-questions using one or more of but not limited to coreference resolution, dependency parsing, chunking and prompt techniques. The method may further include generating code scripts for each sub-question using a Large Language Model (LLM). The method may further include executing the code scripts to retrieve relevant structured data from a plurality of tables. The method may further include generating an answer to the complex question based on the retrieved structured data.