Label-Based Similarity Matrix Visualization for Data Quality
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing data measurement techniques fail to account for variations caused by factors unrelated to the original measurement purpose, such as measurer or measurement site, leading to degraded data quality and hindering accurate identification of target labels.
Innovation Solution
An information processing apparatus and method that generates a similarity matrix, adjusts its order based on label information, and displays an evaluation screen image to visually represent data variation caused by specific labels, allowing for intuitive identification of data quality issues.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If data are measured and utilized without considering variation factors, then measurement simplicity is maintained, but data quality degrades due to variation from factors like measurer or measurement site
Solution Approach 1:
The patent segments the data analysis process by creating separate similarity matrices for different factors (measurer, measurement site, time) and then combining them. This allows the system to account for multiple variation sources without requiring a completely complex new measurement system, as each factor is analyzed independently and then integrated.
Solution Approach 2:
The patent introduces similarity matrices as intermediary computational structures that mediate between raw measurement data and final quality assessment. These matrices serve as intermediate representations that capture relationships between samples while accounting for different variation factors, enabling systematic quality evaluation without direct complex measurement interventions.
2Loss of information
If traditional correlation analysis is used to display data similarity, then time-based variation is visible, but factors other than time (measurer, apparatus) cannot be analyzed
Solution Approach 1:
The patent creates a universal similarity matrix framework that can analyze multiple types of factors (time, measurer, measurement site, apparatus) using the same computational approach. This multi-functional system allows the analysis of diverse variation sources without requiring separate specialized analysis tools for each factor type.
Solution Approach 2:
The patent extends the analysis from traditional one-dimensional time-based correlation to multi-dimensional analysis by incorporating additional factors (measurer, measurement site, apparatus) as separate dimensions. Each factor generates its own similarity matrix, allowing comprehensive variation analysis across multiple dimensions simultaneously.
3Loss of information
If similarity matrices are generated for multiple factors, then comprehensive data variation analysis is achieved, but computational complexity increases
Solution Approach 1:
The computational workload is segmented by calculating similarity matrices for each factor independently (time, measurer, measurement site, apparatus) rather than computing a single complex matrix. This segmentation allows for more efficient computation and enables parallel processing of different factor analyses.
Solution Approach 2:
The system computes similarity matrices for all possible factor combinations, which may seem excessive, but this comprehensive approach actually simplifies the overall analysis by providing complete variation information upfront. The exhaustive computation avoids the need for iterative or approximation methods that could require more computational resources in the long run.
Data Source
AI summary
An information processing apparatus includes: the distance-matrix calculating unit generates, from a data set including samples, a similarity matrix having columns and rows arranged in accordance with an order of the arranged samples, and stores a similarity between a sample corresponding to one of the columns and a sample corresponding to one of the rows in a field specified by the one of the columns and the one of the rows; the order adjusting unit generates an adjusted similarity matrix by referring to label information indicating a result of labeling each of the samples with labels each having classes and adjusting the order in the similarity matrix so that the samples are arranged by each of the classes in a target label designating in the labels; the visualizing unit generates an evaluation screen image indicating each field of the adjusted similarity matrix in brightness corresponding to the similarity.