Neural Framework for Component Problem Extraction

Solution Overview

Problem

Existing knowledge base construction methods for machine learning implementations do not effectively identify component names and their associated problems in text, relying on seed head nouns and requiring significant domain knowledge, which limits their applicability and accuracy.

Innovation Solution

A framework that uses domain-specific syntactic rules and a neural attention-based sequence-to-sequence model to extract domain-specific components and problems from textual corpora, such as service manuals and question-and-answer forums, allowing for unsupervised curation and the creation of a hierarchical knowledge base with 'supertype-subtype' relationships, reducing the need for seed nouns and human supervision.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If seed head nouns and linguistic analysis are used to extract component names, then the extraction process can identify some components, but the coverage is limited and requires significant domain knowledge

Engineering Contradiction:

Improvecomponent extraction coverageVSAvoiddomain knowledge requirement

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The system performs self-service by automatically learning component names and problem types from unannotated text data through unsupervised learning algorithms. The neural model autonomously identifies patterns and entities without requiring manual seed nouns or expert domain knowledge, enabling the system to adapt to new domains independently.

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The patent replaces the mechanical/manual process of domain expert annotation and seed noun selection with an automated neural network system. The deep learning model substitutes human expertise by automatically learning from raw text, eliminating the need for manual domain knowledge injection and significantly reducing complexity.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Reliability

If manual curation and human supervision are used to construct knowledge base, then the knowledge base quality can be maintained, but the construction process is time-consuming and requires human resources

Engineering Contradiction:

Improveknowledge base qualityVSAvoidknowledge base construction speed

Core Design Contradiction:

ReliabilityVSProductivity

Solution Approach 1:

The system performs self-service by automatically learning component names and problem types from unannotated text data through unsupervised learning algorithms. The neural model autonomously identifies patterns and entities without requiring manual seed nouns or expert domain knowledge, enabling the system to adapt to new domains independently.

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The patent replaces the mechanical/manual process of domain expert annotation and seed noun selection with an automated neural network system. The deep learning model substitutes human expertise by automatically learning from raw text, eliminating the need for manual domain knowledge injection and significantly reducing complexity.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

3Adaptability or versatility

If general purpose knowledge base construction methods are used, then the system can handle various topics, but the accuracy of component and problem identification is insufficient

Engineering Contradiction:

Improvetopic coverageVSAvoidcomponent and problem identification accuracy

Core Design Contradiction:

Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent applies local quality by training the neural model on domain-specific text data, allowing it to develop specialized understanding of particular domains (e.g., automotive, electronics). The model adapts its extraction capabilities to local domain characteristics while maintaining the general framework, achieving both versatility and precision through domain-adaptive training.

Inventive Principle:

Principle #3Local quality

Data Source

PatentUS11031009B2Method for creating a knowledge base of components and their problems from short text utterances

Publication Date: 2021.06.08 HITACHI LTD

AI summary

Example implementations involve a framework for knowledge base construction of components and problems in short texts. The framework extracts domain-specific components and problems from textual corpora such as service manuals, repair records, and public Q/A forums using: 1) domain-specific syntactic rules leveraging part of speech tagging (POS), and 2) a neural attention-based seq2seq model which tags raw sentences end-to-end identifying components and their associated problems. Once acquired, this knowledge can be leveraged to accelerate the development and deployment of intelligent conversational assistants for various industrial AI scenarios (e.g., repair recommendation, operations, and so on) through better understanding of user utterances. The example implementations give better tagging accuracy on various datasets outperforming well known off-the-shelf systems.