Swarm Systems in the Visualization of Consumption Patterns

Catarina Maçãs, Pedro Cruz, Pedro Martins, Penousal MachadoIJCAI 2015

Introduction

This research, originally presented at IJCAI 2015, explores the intersection of Information Aesthetics and Multi-Agent Systems (MAS). The authors demonstrate how swarm-based simulations can transform massive, dry datasets—specifically two years of retail consumption records—into organic, engaging visualizations that reveal temporal patterns and anomalies. By mapping consumption values to agent behaviors and radial positions on a spiral timeline, they created "living" visualizations that balance artistic expression with the functional detection of periodic retail trends.

Problem Statement

Traditional data visualization often struggles with "Big Data" in two ways: it can be visually overwhelming, or it can be too sterile to engage a non-expert audience. The authors identify a gap in Information Aesthetics, a sub-field that sits between functional data visualization and fine art. The challenge is to create a system that can handle dynamic, time-varying datasets, readjust continuously to data changes, and lure viewers into "decoding" the information through aesthetic appeal. Specifically, they tackle the visualization of 289 GB of consumption data, looking for a way to show not just the volume of sales, but the rhythmic nature of consumption over time.

Methodology

The authors implement a Swarm System based on Craig Reynolds’ classic "Boids" model. In this system, each "boid" represents one of seven product departments (e.g., Grocery, Leisure, Fresh Food). These agents follow three standard steering behaviors: cohesion (staying near peers), separation (avoiding collisions), and alignment (matching velocity).

To represent the temporal dimension, the authors introduced a "target boid" that moves in a spiral from the center of the canvas outward. All other boids are attracted to this target, creating a chronological flow where one full lap of the spiral represents one month. Data is aggregated into two-hour intervals, resulting in 12 intervals per day.

The spiral timeline mapping, where one lap represents a month and radial positions correspond to days.

Data is mapped to visual attributes as follows:

  • Radius: The size of a boid (and its corresponding separation force) is directly proportional to the consumption volume at that time interval.
  • Color: Assigned by department to distinguish product categories.
  • Rendering: The authors experimented with various styles, including filled circles, strokes, and lines connecting boids of the same department to highlight clusters.

Dataset

The study uses a real-world dataset from SONAE, the largest retail chain in Portugal. The scope includes 729 hypermarkets and supermarkets over a duration from May 2012 to April 2014. This scale involves approximately 289 GB of transaction data across seven departments: Grocery, Fresh Food, Food & Bakery, Home, Leisure, Textile, and Health.

Results and Discussion

The system produces emergent ordering: because larger consumption values result in larger boid radii and stronger separation forces, boids with high sales naturally push to the "outside" of the spiral path. This creates a visual hierarchy where peak shopping periods literally "burst" outward, making the "heartbeat" of retail—daily cycles, weekend surges, and seasonal anomalies like the Christmas rush—clearly visible.

Comparison of different rendering techniques, from discrete circles to continuous lines, illustrating the trade-off between readability and artistic expression.

The authors explored different normalization techniques to reveal various insights. Global normalization makes it easy to see which departments dominate (typically Grocery and Fresh Food), while per-department normalization allows the viewer to see internal rhythms within smaller departments, such as a sudden spike in the "Leisure" department during annual toy promotions.

Regarding aesthetic engagement, the resulting images resemble organic "digital landscapes." However, a trade-off exists between readability and art. While connecting boids with lines created beautiful, complex "tangles," it often made the data harder to interpret. The most "readable" version used sorted, filled circles that prevented smaller data points from being hidden by larger ones.

Conclusion

As an exploration of Information Aesthetics, the success of this visualization is partly subjective, depending on the viewer's willingness to engage with the visual legend. While some artistic renderings sacrifice functional clarity for visual impact, the system successfully highlights atypical consumption patterns through emergent behavior. Future work will involve formal user evaluations to determine how effectively non-experts can "read" these swarm patterns and the exploration of new behavioral rules to convey data patterns even more efficiently.

Summary updated

BibTeX
@inproceedings{macas2015swarm,
  author = {Catarina Maçãs and Pedro Cruz and Pedro Martins and Penousal Machado},
  title = {{Swarm Systems in the Visualization of Consumption Patterns}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {2015},
}
Figure