Neural Network Question Vector Search for Answer Retrieval

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing methods for determining similar questions in large text document databases are inefficient, leading to increased search times as the database size grows, as they require searching through a large amount of text to find corresponding answers.

Innovation Solution

A neural network-based similarity analysis model is used to generate user question vectors, select similar questions based on similarity analysis, and determine answers from these similar questions, reducing the need to search through the entire database by focusing on questions with higher similarity scores.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If a large amount of text document database is searched to find answers to user questions, then the completeness of answer coverage is improved, but the search time is increased

Engineering Contradiction:

Improveanswer coverage completenessVSAvoidsearch time

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The system pre-generates question vectors for all documents in the database before actual search operations. This preliminary vectorization allows the search to proceed by comparing user query vectors against pre-computed document vectors, significantly reducing search time while maintaining complete answer coverage across the entire database.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent replaces traditional text-based search mechanisms with neural network-based vector similarity comparison. Instead of searching through text documents directly, the system converts both user queries and documents into vector representations and uses vector similarity metrics to identify relevant documents, dramatically improving search efficiency.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Adaptability or versatility

If traditional text-based similarity analysis is used, then the system can handle diverse question formats, but the search efficiency decreases as database size increases

Engineering Contradiction:

Improvequestion format handlingVSAvoidsearch efficiency

Core Design Contradiction:

Adaptability or versatilityVSProductivity

Solution Approach 1:

The system transforms questions from text format into vector space representations using neural networks. This parameter transformation allows the system to handle diverse question formats uniformly by converting them all into vectors, while enabling efficient similarity computation through vector operations that scale better with database size.

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent substitutes traditional text-based similarity analysis with neural network-based vector similarity analysis. This replacement maintains the ability to handle diverse question formats while improving search efficiency, as vector comparison operations are computationally more efficient than text processing, especially for large databases.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentUS11288265B2Method and apparatus for building a paraphrasing model for question-answering

Publication Date: 2022.03.29 42 MARU INC

AI summary

Disclosed herein is a search method including: generating a user question vector for a user question using a paraphrase model learned using first learning data composed of a first pair of questions and a label indicating that the first pair of questions are similar to each other and second learning data composed of a second pair of questions and a label indicating that the second pair of questions are dissimilar to each other; selecting a similar question based on a similarity analysis result to the user question vector; and determining an answer to the similar question as an answer to the user question.