Integrating topic modeling and word embedding to characterize violent deaths.

Area of Science:

• Computational linguistics
• Public health research
• Data science

Background:

• Administrative databases contain rich textual data.
• Analyzing large-scale text data presents computational challenges.
• Existing classification schemes for violent death may miss nuanced aspects.

Purpose of the Study:

• To develop and apply an integrated computational approach for identifying latent topics in large-scale text data.
• To analyze narratives of violent death from the National Violent Death Reporting System (NVDRS).
• To uncover aspects of violent death not captured by current classifications and identify potential gender biases.

Main Methods:

• Integration of topic modeling and word embedding techniques for computational text analysis.
• Application of the developed approach to the National Violent Death Reporting System (NVDRS) dataset.
• Extraction and analysis of latent topics and associated gender biases within the text data.

Main Results:

• Identification of novel topics within violent death narratives, revealing previously uncaptured aspects.
• Discovery of gender bias within specific topics, such as a "long guns" topic being associated with masculinity.
• Demonstration of the approach's ability to extract nuanced information beyond existing classification schemes.

Conclusions:

• The integrated topic modeling and word embedding approach effectively identifies latent topics and biases in large text datasets.
• Findings offer new research directions for suicide and homicide prevention.
• The methodology is broadly applicable to diverse administrative databases for unlocking similar insights.