Natural Language Query Translation via Ensemble Classifiers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current systems face challenges in translating natural language queries into structured queries, particularly for non-technical users, due to inadequate comprehension of user intent and rigidity in system configuration, leading to limited accessibility of structured data.
Innovation Solution
The method involves using multiple natural language classifiers trained with different ground truths to determine candidate question types, assigning weights based on query focus, and selecting the most confident structured question type, with the option to request clarification if confidence levels are below a threshold, and inserting specific data into structured data variables to form a structured data query.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single natural language classifier is used to translate natural language queries into structured queries, then the system configuration is simple, but the accuracy of understanding user intent and determining the correct structured question type is insufficient
Solution Approach 1:
The patent combines multiple natural language classifiers (NLCs) with different ground truths into a single system that processes natural language queries. Each NLC is trained on different aspects or domains of structured data, and their outputs are aggregated to determine the final structured question type. This merging approach improves intent understanding accuracy by leveraging the strengths of multiple specialized classifiers rather than relying on a single general-purpose classifier.
Solution Approach 2:
The system employs multiple NLCs that can each handle different aspects of query translation independently. These classifiers are designed with universal applicability through a common framework that accepts various ground truths and produces structured question types. The system can switch between or combine different NLCs based on the query content, making the overall system multi-functional and adaptable to diverse query types while maintaining high accuracy for each specific domain.
2Reliability
If multiple natural language classifiers with different ground truths are used to determine candidate question types, then the accuracy of translating natural language to structured queries is improved, but the system complexity and computational resources required increase
Solution Approach 1:
The patent segments the query translation task into multiple independent classification sub-tasks, each handled by a specialized NLC trained on specific ground truths. Instead of using one large complex classifier, the system divides the problem into manageable segments where each NLC focuses on a particular aspect or domain. The results from these segmented classifications are then integrated to produce the final structured question type, improving reliability while keeping individual classifier complexity manageable.
Solution Approach 2:
The system dynamically adjusts parameters such as the number of NLCs to deploy, the weightings assigned to each classifier's output, and the confidence thresholds for selecting structured question types. These parameters can be changed based on the specific query, the available computational resources, and the required level of accuracy. This flexibility allows the system to optimize between reliability and complexity by adjusting parameters rather than fixedly using a predetermined number of classifiers.
Data Source
AI summary
A natural language query (NLQ) is translated to a structured data query (e.g., a SQL statement) by extracting entities from the NLQ and replacing them with generic variables to form a generic query. The generic query is associated with a structured question type which includes structured data variables using natural language classifiers (NLCs). Specific data is inserted in the structured question type in relation to the structured data variables based on the extracted entities to form the structured data query. An ensemble of NLCs trained with different ground truths can be used to yield multiple candidate question types. One of the candidate question types is selected based on confidence levels. The multiple NLCs can include an NLC which is optimized according to a focus of the generic query. For example, an NLC can be optimized for a specific data structure (such as SQL), or for comparative queries.