Visually-supported topic modeling for understanding behavioral patterns from spatio–temporal events

Spatio-temporal event sequences consist of activities or occurrences involving various interconnected elements in space and time. We show how topic modeling—typically used in text analysis—can be adapted to abstract and conceptualize such data. We propose an overall analytical workflow that combines computational and visual analytics methods to support some tasks, enabling the transformation of raw event data into meaningful insights. We apply our workflow to football matches as an example of important yet under-explored spatio-temporal event data. A key step in topic modeling is determining the appropriate number of topics; to address this, we introduce a visual method that organizes multiple modeling runs into a similarity-based layout, helping analysts identify patterns that balance interpretability and granularity. We demonstrate how our workflow, which integrates visual analytics, supports five core analysis tasks: identifying common behavioral patterns, tracking their distribution across individuals or groups, observing progression at different temporal scales, comparing behavior under varied conditions, and detecting deviations from typical behavior. Using real-world football data, we illustrate how our end-to-end process enables deeper insights into both tactical details and broader trends — from single match analyses to season wide perspectives. While our case study focuses on football, the proposed workflow is domain-agnostic and can be readily applied to other spatio-temporal event datasets, offering a flexible foundation for extracting and interpreting complex behavioral patterns. • An end-to-end workflow for transforming raw spatio-temporal events into meaningful insights. • A visual analytics technique for selecting the number of topics with the most suitable result from multiple runs of topic modeling on spatial event sequences. • A demonstration of how our technique can be used to discover spatially consistent and easily distinguishable topics from a football dataset. • A demonstration of how the obtained topics can be used to summarize a football game’s dynamics and reveal insights into the games. • Methods for using the topics to summarize and compare various matches and teams’ playing styles. • An exploration of how different teams deviate from their typical play style in different conditions.