AI Response Verification Using Inquiry-Answer Classification Matching

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing natural language processing systems using generative AI often generate answers irrelevant to user inquiries, particularly in models that reduce data bit width for high-speed text generation, leading to inferior inference accuracy.

Innovation Solution

An information processing device and method that utilizes a classification model to compare the classification of user inquiries and AI-generated responses, regenerating prompts until a matching classification is achieved, ensuring relevant and accurate answers are provided.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Speed

If data bit width is reduced for high-speed text generation, then text generation speed is improved, but inference accuracy deteriorates

Engineering Contradiction:

Improvetext generation speedVSAvoidinference accuracy

Core Design Contradiction:

SpeedVSMeasurement precision

Solution Approach 1:

A classification model is introduced as an intermediary component between the generative AI and the user. This classification model verifies whether generated answers are relevant to user inquiries by comparing classifications of questions and answers, acting as a mediator that ensures accuracy without requiring the main generative model to be more complex or slower.

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The system is divided into separate functional components: a generative AI model for producing text and a classification model for verifying relevance. This segmentation allows each component to be optimized independently - the generative model can use reduced bit width for speed while the classification model ensures accuracy through separate verification.

Inventive Principle:

Principle #1Segmentation

2Measurement precision

If a classification verification step is added to ensure answer relevance, then answer accuracy is improved, but processing time increases

Engineering Contradiction:

Improveanswer accuracyVSAvoidprocessing time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The system replaces manual verification of answer relevance with an automated classification model that uses machine learning to quickly determine whether an answer is relevant to a user inquiry. This substitution of mechanical/manual verification with automated computational classification reduces the time penalty while maintaining accuracy improvements.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentUS20260057188A1Information processing device, information processing method, and information processing system

Publication Date: 2026.02.26 TOSHIBA TEC KK

AI summary

According to one embodiment, an information processing device includes a communication interface and a control unit that receives an inquiry sentence from a user, classifies the inquiry sentence by submitting the inquiry sentence to a classification model, generates a prompt from the inquiry sentence, then inputs the prompt to a generative AI to receive a response sentence. The control unit then classifies the response sentence by submitting the response sentence to the classification model, then determines whether the inquiry sentence classification matches the response sentence classification. The control unit outputs the response sentence to the user via the communication interface when the classifications are determined to be similar, but generates another prompt from the inquiry sentence to be input to the generative AI when the classifications are dissimilar.