Search System Using Classification-Based Database Segmentation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing search systems using learners to find similar information are limited in accuracy as they can only search for similarities based on known features learned by the system, failing to account for unknown features.

Innovation Solution

A search system that divides databases into classifications based on the analysis results of input information, allowing for the search of information similar to the input within the corresponding classification, thereby improving search accuracy by utilizing feature vectors and scores as indexes.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If the learner searches for similar information based on known features learned in the system, then the search can be performed using existing analysis results, but the accuracy of the search cannot be fully improved because unknown features are not considered

Engineering Contradiction:
Improvesearch accuracyVSAvoidability to detect unknown features
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent segments the search process into two distinct phases: (1) classification-based filtering that uses known features to divide the database into relevant categories, and (2) similarity search within those categories that can detect both known and unknown features. This segmentation allows the system to leverage the strengths of both supervised classification and unsupervised similarity measurement.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces an intermediary mechanism (the classification result based on known features) that mediates between the input information and the final similarity search. This intermediary filters and organizes the search space, enabling more effective detection of similarities including unknown features within each classification category.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Measurement precision

If the system searches all information in the database without classification, then it can potentially find similar information based on any features, but the search process becomes inefficient and time-consuming

Engineering Contradiction:
Improvecomprehensive similarity detectionVSAvoidsearch speed
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent applies segmentation by dividing the entire database into multiple classification categories based on known features before performing similarity searches. This reduces the search space from the entire database to only relevant subsets, dramatically improving search speed while maintaining comprehensive similarity detection within each category.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent performs preliminary classification of the database based on known features before conducting the actual similarity search. This preliminary action organizes and filters the data in advance, so that when the similarity search is executed, it operates on a pre-filtered subset rather than the entire database, improving efficiency.

Inventive Principle:
Principle #10Preliminary action

3Productivity

If the learner uses only classification results for search, then the search is fast and simple, but it cannot detect similarities in features that were not learned during training

Engineering Contradiction:
Improvesearch efficiencyVSAvoiddetection of unknown features
Core Design Contradiction:
ProductivityVSMeasurement precision

Solution Approach 1:

The patent segments the feature detection capability into two layers: the classification layer that handles known features efficiently, and the similarity search layer that handles both known and unknown features. This segmentation allows each layer to specialize in its strength while working together to achieve comprehensive search capability.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent adds another dimension to the search process by introducing similarity search as a complementary approach to classification-based search. While classification operates in the dimension of known labeled features, similarity search operates in the broader dimension of feature space including unlabeled and unknown features, providing comprehensive coverage.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

Data Source

PatentEP3751424B1Search system, search method, and program
Publication Date: 2025.05.14 RAKUTEN GROUP INC
  • EP3751424B1 patent drawingFigure 1
  • EP3751424B1 patent drawingFigure 2
  • EP3751424B1 patent drawingFigure 3

AI summary

To improve accuracy of search, a learner (L) of a search system (1) calculates a feature quantity of information that is input and outputs a first analysis result of the information in a first viewpoint and a second analysis result of the information in a second viewpoint based on the feature quantity. Storing means (102) stores a feature quantity of information to be searched, which has been input in the learner (L), in a database. Input means (103) inputs input information in the learner (L). Search means (104) searches for information to be searched that is similar to the input information in the feature quantity based on the database.