Word Similarity Extraction via User-Contextualized Distributed Representations

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing conference coordination systems require users to specify multiple reservation conditions manually, leading to user burden and inefficiency, and struggle to accurately extract words with similar concepts due to noise in similarity calculations based on simple distributed representation comparisons.

Innovation Solution

An information processing device that calculates distributed representations of character strings combining user names and words, then uses similarity calculations and normalization to extract words with similar concepts by analyzing the frequency and rank of these representations.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Device complexity

If simple distributed representation comparison is used for word similarity calculation, then the calculation process is simple, but noise in similarity calculations increases and detection accuracy decreases

Engineering Contradiction:
Improvecalculation process complexityVSAvoidword similarity detection accuracy
Core Design Contradiction:
Device complexityVSMeasurement precision

Solution Approach 1:

The patent segments the word similarity calculation process into multiple stages: initial distributed representation calculation, similarity score computation, rank determination, and frequency-based noise filtering. This segmentation allows each stage to address specific aspects of the problem, improving overall accuracy while maintaining manageable complexity

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces intermediate variables (similarity scores, ranks, and frequency counts) as mediators between the raw distributed representations and the final similarity determination. These intermediaries enable systematic noise filtering and multi-criteria evaluation, resolving the contradiction between simple calculation and accurate detection

Inventive Principle:
Principle #24Intermediary (Mediator)

2Measurement precision

If multiple reservation conditions are specified manually in existing conference coordination systems, then conference room search accuracy is maintained, but user burden increases and efficiency decreases

Engineering Contradiction:
Improveconference room search accuracyVSAvoiduser operation burden
Core Design Contradiction:
Measurement precisionVSEase of operation

Solution Approach 1:

The system performs self-service by automatically calculating word similarities and extracting relevant keywords without requiring users to manually specify multiple reservation conditions. The automated similarity calculation and keyword extraction mechanisms enable the system to independently determine appropriate conference rooms based on minimal user input

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The patent implements preliminary action by pre-calculating distributed representations and similarity metrics for potential conference room keywords before user queries. This preprocessing enables rapid response to user inputs while maintaining search accuracy, eliminating the need for users to manually specify multiple conditions

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentUS11893814B2Extraction method, computer-readable recording medium storing extraction program, and information processing device
Publication Date: 2024.02.06 FUJITSU LTD
  • US11893814B2 patent drawing
  • US11893814B2 patent drawing
  • US11893814B2 patent drawing

AI summary

An extraction method executed by a computer includes: acquiring a first word input by a first user; acquiring a second word input by a second user; calculating a first distributed representation of a first character string, the first character string including the first word and a character string that identifies the first user; calculating a second distributed representation of a second character string, the second character string including the second word and a character string that identifies the second user; calculating a similarity between the first distributed representation and the second distributed representation; and extracting a word associated with the second distributed representation on a basis of the similarity.