A large model hallucination suppression intelligent question and answer system and method based on deep learning

By improving NASNet network and database structure awareness technology, and combining it with consistency verification, the illusion problem in intelligent question answering systems has been solved, improving the accuracy and reliability of query results and ensuring the legality and consistency of generated answers.

CN121722786BActive Publication Date: 2026-05-26KEXUN JIALIAN INFORMATION TECH CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
KEXUN JIALIAN INFORMATION TECH CO LTD
Filing Date
2026-02-11
Publication Date
2026-05-26

AI Technical Summary

Technical Problem

Existing deep learning-based large language models are prone to producing illusions in intelligent question-answering systems, generating answers that do not conform to the facts, affecting the accuracy and reliability of the system, especially in database query and knowledge reasoning scenarios.

Method used

By improving NASNet network, large model inference, database structure awareness, and consistency verification techniques, a database schema is constructed, task prompt word sequences are configured, SQL query statements are generated, and legality and consistency verification is performed to ensure that the query results match the user's intent.

Benefits of technology

It significantly improves the accuracy and reliability of query results, avoids query failures due to structural mismatches or syntax errors, and enhances the system's query comprehension and the accuracy of answers.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121722786B_ABST
    Figure CN121722786B_ABST
Patent Text Reader

Abstract

This invention discloses a deep learning-based intelligent question-answering system and method for suppressing illusions in large models, comprising the following steps: constructing a database schema and configuring entity mapping relationships; configuring prompt word information for large model inference to form a task prompt word sequence; forming a complete system input; using a large model with an improved NASNet network architecture to perform structured query language transformation and generate SQL query statements; generating natural language description information; performing consistency verification and obtaining consistency verification results; generating corresponding computational prompt information; obtaining the query results, and outputting the query results and computational prompt information together as intelligent question-answering results. This invention effectively suppresses illusions and semantic shifts generated by large models in structured question-answering scenarios, improving the accuracy of SQL generation and the reliability of question-answering results.
Need to check novelty before this filing date? Find Prior Art