Intelligent Response Agent With Evidential Reasoning for Deep Document QA

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing deep learning-based question-answering systems struggle with providing accurate responses to in-depth queries requiring sophisticated reasoning and inference, particularly in specialized documents like scientific and technical papers, due to the exponential increase in syntactic patterns and the lack of training on deep document comprehension.

Innovation Solution

A method and system using a deep-learning neural network that implements a stepwise process for sophisticated reasoning and inference, including acquiring query data, detecting evidential paragraphs, generating response data, and providing rationale information, to enhance the accuracy and reliability of responses.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional question-answering systems construct semantic elements as patterns by parsing queries, then they can extract responses from database structures, but the number of syntactic patterns increases exponentially causing decreased effectiveness

Engineering Contradiction:
Improveanswer accuracyVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent replaces the mechanical pattern-matching system with a neural network-based semantic understanding system. Instead of constructing and matching syntactic patterns, the system uses natural language processing models to comprehend query intent and generate answers directly, eliminating the exponential growth of syntactic patterns while maintaining answer accuracy.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent changes the fundamental parameter of query processing from syntactic pattern matching to semantic understanding. By transforming the approach from rule-based pattern construction to neural network-based meaning comprehension, the system achieves accurate answers without the complexity explosion inherent in traditional methods.

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If MRC models are used to analyze queries and find optimized answers, then answer quality improves, but a presented text must first exist making it difficult to provide responses when no text has been generated

Engineering Contradiction:
Improveanswer qualityVSAvoidresponse capability
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent creates a universal question-answering system that can handle both retrieval-based scenarios (where source text exists) and generative scenarios (where no source text is available). The neural network model is trained to perform both functions, making the system adaptable to different contexts without requiring separate processing pipelines.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The patent implements a dynamic system that adapts its behavior based on the availability of source text. When text is available, the model leverages it for grounded answers; when text is unavailable, the model transitions to generating answers from its learned knowledge, making the system flexible across different operational conditions.

Inventive Principle:
Principle #15Dynamics

3Measurement precision

If MRC models are trained on question-answering datasets, then basic factoid QA capability is enhanced, but deep document comprehension and sophisticated reasoning remain largely undeveloped

Engineering Contradiction:
Improvefactoid QA accuracyVSAvoiddeep reasoning capability
Core Design Contradiction:
Measurement precisionVSReliability

Solution Approach 1:

The patent performs preliminary training on extensive question-answering datasets to establish basic factoid QA capability, then builds upon this foundation with additional training focused on deep document comprehension and reasoning tasks. This staged approach ensures both basic accuracy and advanced reasoning capability are developed.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent creates a composite training approach that combines multiple dataset types and training objectives. By integrating factoid QA training with deep comprehension training on specialized documents, the system develops both basic answer accuracy and sophisticated reasoning capabilities in a unified model.

Inventive Principle:
Principle #40Composite materials

4Adaptability or versatility

If deep learning models process in-depth queries requiring knowledge processing and reasoning, then comprehensive answers can be provided, but answer accuracy and quality remain low for specialized documents

Engineering Contradiction:
Improvequery handling capabilityVSAvoidanswer accuracy
Core Design Contradiction:
Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent applies local quality by training the model specifically on specialized documents in particular domains (e.g., scientific papers, technical manuals). This domain-specific training enhances the model's ability to provide accurate answers for specialized queries while maintaining general capabilities for other types of questions.

Inventive Principle:
Principle #3Local quality

Data Source

PatentEP4679287A1Method for providing intelligent response agent based on advanced inference and estimation function, and system therefor
Publication Date: 2026.01.14 LG MANAGEMENT DEV INST CO LTD
  • EP4679287A1 patent drawingFigure 1
  • EP4679287A1 patent drawingFigure 2~3
  • EP4679287A1 patent drawingFigure 4~5

AI summary

A method and system for providing an intelligent response agent based on a sophisticated reasoning and speculation function according to an embodiment of the present disclosure can generate and provide response data for queries related to specialized documents using a deep-learning neural network that implements a stepwise process for a sophisticated reasoning and speculation function.