LLM-Generated QnA Pairs for Q&A Vector Search Accuracy

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Vector search algorithms in QnA platforms struggle to find similar questions that are not directly asked, leading to inaccurate results due to limitations in vector embeddings of text data.

Innovation Solution

Employing generative artificial intelligence, such as large language models, to generate additional QnA pairs from existing datasets, which are then embedded in a vector space for enhanced search accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If vector search algorithms are used to perform similarity searches based on user queries, then the QnA platform can automatically retrieve relevant information, but the search accuracy deteriorates when users ask questions that are not directly like those already in the database

Engineering Contradiction:

Improveautomatic information retrieval capabilityVSAvoidsearch accuracy

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

The system performs preliminary actions by generating synthetic QnA pairs before the actual search operation. Large language models create additional question-answer pairs from existing data, enriching the vector database proactively so that when user queries arrive, the search algorithm has more relevant references to work with, improving accuracy without sacrificing automation

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The invention creates copies of existing QnA pairs through synthetic data generation. By using large language models to generate multiple variations and related questions from original QnA pairs, the system multiplies the effective content in the vector database, allowing more accurate matching of user queries even when they differ from original formulations

Inventive Principle:

Principle #26Copying

2Reliability

If the vector database contains only original QnA pairs from external sources, then the data quality is maintained, but the search coverage and answer space density are limited

Engineering Contradiction:

Improvedata qualityVSAvoidsearch coverage

Core Design Contradiction:

ReliabilityVSAdaptability or versatility

Solution Approach 1:

The system changes the parameter of data quantity and variety by generating synthetic QnA pairs. Large language models create additional entries with varied formulations, questions, and answers based on original data, expanding the vector database's coverage and adaptability while maintaining data quality through controlled generation processes

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS12363054B1Systems and methods for enhancing vector searches for question-and-answer platforms

Publication Date: 2025.07.15 INTUIT INC

AI summary

Systems and methods are provided for enhancing vector searches for question-and-answer platforms.