Visualization of Large-Scale Narrative Data Describing Human Error.

Area of Science:

• Data Science
• Information Visualization
• Natural Language Processing

Background:

• Analyzing large datasets presents significant challenges.
• Visual representation is a powerful tool for data analysis.
• Existing methods struggle with the scale and complexity of narrative data.

Purpose of the Study:

• To develop and evaluate a visual data exploration technique for text-based narratives.
• To assess the utility of latent semantic analysis (LSA) combined with Geographic Information Systems (GIS) for data reduction and organization.
• To improve the analysis of large narrative datasets in domains like aviation safety.

Main Methods:

• Utilizes latent semantic analysis (LSA) to reduce narratives into high-dimensional vectors.
• Applies isometric mapping to truncate vectors into two-dimensional projections.
• Visually renders the projected data using Geographic Information Systems (GIS) software.
• Demonstrates the technique on aviation self-reported safety narratives from the Aviation Safety Reporting System.

Main Results:

• Successfully assimilates and organizes a large corpus of aviation safety narratives.
• Identifies distinct thematic regions and the first five major topics within the data.
• Illustrates the ability to recognize common events and identify outliers.
• Provides a visual representation of narrative data for enhanced comprehension.

Conclusions:

• The LSA-based projection into GIS is an effective tool for reducing and organizing large narrative datasets.
• This visual data exploration technique facilitates efficient analysis, comprehension, and safety improvements in aviation and other domains.
• Enables identification of contextual themes, common events, and outliers within narrative data.