Graph Data Interpretation via Alternative Sample Retrieval

Solution Overview

Problem

Users face difficulties in interpreting the basis of determination results for graph data, particularly when scores corresponding to all elements of the adjacency matrix are relatively small, making it challenging to distinguish between the absence of a topological feature and poor accuracy of the GXAI model.

Innovation Solution

A determination method that acquires scores for graph data using a trained machine learning model and refers to a storage device to specify alternative graph data with different determination results when all initial scores are below a threshold, facilitating the interpretation of the basis of the determination result by outputting information associated with the specified graph data.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a trained machine learning model outputs scores for graph data elements, then the determination result can be obtained, but when all scores are below the threshold, it becomes difficult to interpret whether the topological feature is truly absent or the model accuracy is poor

Engineering Contradiction:

Improveinterpretation accuracyVSAvoidbasis information

Core Design Contradiction:

Measurement precisionVSLoss of information

Solution Approach 1:

The system performs preliminary action by storing multiple graph data samples with their determination results in advance. When interpretation is needed, these pre-stored samples are retrieved and presented to the user, eliminating the need for real-time analysis of alternative scenarios and providing immediate interpretative context.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system introduces an intermediary mechanism by presenting alternative graph data samples that serve as mediators between the model's determination result and the user's understanding. These samples act as reference points that help users interpret the basis of determination results when score explanations are insufficient.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Loss of information

If the system presents score information for all elements of the adjacency matrix, then the basis of determination is provided, but users still face difficulty in interpreting the meaning when all scores are small

Engineering Contradiction:

Improvebasis informationVSAvoidinterpretability

Core Design Contradiction:

Loss of informationVSEase of operation

Solution Approach 1:

The system performs preliminary action by storing multiple graph data samples with their determination results in advance. When interpretation is needed, these pre-stored samples are retrieved and presented to the user, eliminating the need for real-time analysis of alternative scenarios and providing immediate interpretative context.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

Instead of directly explaining why scores are low (which is difficult), the system inverts the approach by presenting alternative graph data samples that have different determination results. Users can interpret the basis by comparing the original data with these alternative samples, making the interpretation process more intuitive.

Inventive Principle:

Principle #13The other way round (Inversion)

3Ease of operation

If the system outputs only the determination result without additional information, then the output is simple, but users cannot understand the basis of the determination when scores are low

Engineering Contradiction:

Improveoutput simplicityVSAvoidinterpretation basis

Core Design Contradiction:

Ease of operationVSLoss of information

Solution Approach 1:

The system performs preliminary action by storing multiple graph data samples with their determination results in advance. When interpretation is needed, these pre-stored samples are retrieved and presented to the user, eliminating the need for real-time analysis of alternative scenarios and providing immediate interpretative context.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

Instead of directly explaining why scores are low (which is difficult), the system inverts the approach by presenting alternative graph data samples that have different determination results. Users can interpret the basis by comparing the original data with these alternative samples, making the interpretation process more intuitive.

Inventive Principle:

Principle #13The other way round (Inversion)

Data Source

PatentUS20240119258A1Computer-readable recording medium storing determination program, determination method, and information processing device

Publication Date: 2024.04.11 FUJITSU LTD

AI summary

A computer-readable recording medium storing a program for causing a computer to execute processing including: acquiring a first determination result of first graph data by performing determination processing on the first graph data; acquiring one or more first scores regarding a feature of the first graph data by using a trained model, the one or more first scores representing a basis of the first determination result of the first graph data, the trained model being a model configured to output, in response to obtaining graph data, one or more scores regarding the feature of the graph data; in a case where all of the one or more first scores are less than a threshold, specifying second graph data being a second determination result different from the first determination result; and outputting, in association with the first determination result, information regarding the feature of the second graph data.