Spherical Similarity Explorer for Comparative Case Analysis

Leishi Zhang, Chris Rooney, Lev Nachmanson, B. L. William Wong, Bum Chul Kwon, Florian Stoffel, Michael Hund, Nadeem Qazi, Uchit Singh, Daniel A. Keim

Abstract

Comparative Case Analysis (CCA) is a critical task for criminal investigators, yet it remains largely manual and labor-intensive. This paper introduces the Spherical Similarity Explorer (SSE), a visual analytics system that automates the identification of patterns and links across large-scale crime datasets. By mapping high-dimensional data onto an interactive 3D spherical surface using Multidimensional Scaling (MDS), the SSE provides a continuous, boundary-less environment for exploring similarities in modus operandi (MO), geography, and time. Evaluation with professional analysts demonstrates high satisfaction and significant efficiency gains in pattern discovery compared to traditional 2D visualization methods.

Introduction

In criminal intelligence, identifying commonalities between disparate crime reports is essential for linking cases to a single offender or identifying emerging trends. This process, known as Comparative Case Analysis (CCA), typically involves examining the modus operandi (MO), location, and temporal data of incidents. Historically, analysts have performed this task manually, reading through stacks of reports and collating data into spreadsheets.

As crime data grows in volume and complexity, manual analysis becomes unsustainable. Furthermore, existing visualization tools often rely on 2D scatter plots or "galaxy views," which suffer from "corner effects" and boundary distortions that can misrepresent the relationships between data points. The Spherical Similarity Explorer addresses these challenges by providing an automated pipeline for similarity calculation and an intuitive 3D interface for exploration.

Methods and System Architecture

The SSE system employs a multi-stage visual analytics pipeline to transform raw, unstructured crime reports into an interactive 3D environment.

The visual analytics pipeline for the Spherical Similarity Explorer, from feature extraction to interactive analysis.
  1. Feature Extraction: The system utilizes Natural Language Processing (NLP) techniques, including POS tagging and lemmatization, to extract structured concepts from free-text MO descriptions.
  2. Composite Distance Measure: Similarity between crimes is determined by a weighted combination of three factors: temporal difference, geo-spatial distance, and categorical overlap of MO concepts (calculated via a modified Jaccard approach).
  3. Spherical MDS: To visualize these high-dimensional relationships, the system uses the Cox and Cox algorithm to project the distance matrix onto the surface of a 3D unit sphere. This ensures that similar crimes are clustered together on a continuous surface.
  4. Interactive Linked Views: The 3D sphere serves as the primary navigation tool, dynamically linked to a geo-map for hot spot analysis, a timeline, and a word cloud for identifying common terminology within selected clusters.

Key Insights: The Globe Metaphor

A central innovation of the SSE is the use of spherical geometry. Unlike a 2D plane, a sphere has no edges, which avoids the artificial clustering of data points at the boundaries of a screen. This "globe" metaphor supports the "Focus + Context" principle: as an analyst rotates the sphere to focus on a specific cluster, dissimilar points naturally rotate to the back, reducing visual clutter. Furthermore, the system allows analysts to toggle specific features (Spatial, Temporal, or MO) to test hypotheses and observe how clusters reorganize in real-time.

Key Results and Evaluation

The system was evaluated using 1,588 real-world crime records and feedback from professional analysts in the UK and Belgium. The tool received high scores for satisfaction (4.83/5) and evidential reasoning (4.61/5).

Interactive exploration of crime clusters, showing how the spherical interface links to geo-spatial hot spots and common MO terms like "office."

In one practical use case, analysts identified a distinct cluster on the sphere. By examining the linked word cloud and geo-map, they quickly determined that the cluster represented a series of burglaries targeting non-residential buildings, such as offices and schools. Analysts noted that the system could identify such patterns in minutes, whereas manual cross-referencing of MO traits across thousands of reports would typically take hours or days.

Conclusion and Future Work

The Spherical Similarity Explorer demonstrates the power of 3D spherical projections for high-dimensional data analysis in criminal justice. While the tool was highly praised, analysts suggested improvements such as replacing word clouds with simple frequency lists for easier reporting and providing more transparency regarding the specific features driving similarity scores (the "black box" problem). Future research will focus on dynamic azimuthal projections to allow for absolute distance measurements from specific "anchor" crimes and refining interaction models to avoid the fatigue associated with gesture-based controls.

Summary updated

Abstract

Comparative Case Analysis (CCA) is an important tool for criminal investigation and crime theory extraction. It analyzes the commonalities and differences between a collection of crime reports in order to understand crime patterns and identify abnormal cases. A big challenge of CCA is the data processing and exploration. Traditional manual approach can no longer cope with the increasing volume and complexity of the data. In this paper we introduce a novel visual analytics system, Spherical Similarity Explorer (SSE) that automates the data processing process and provides interactive visualizations to support the data exploration. We illustrate the use of the system with uses cases that involve real world application data and evaluate the system with criminal intelligence analysts.
BibTeX
@inproceedings{zhang2016spherical,
  author = {Leishi Zhang and Chris Rooney and Lev Nachmanson and B. L. William Wong and Bum Chul Kwon and Florian Stoffel and Michael Hund and Nadeem Qazi and Uchit Singh and Daniel A. Keim},
  title = {{Spherical Similarity Explorer for Comparative Case Analysis}},
  booktitle = {Visualization and Data Analysis},
  year = {2016},
  doi = {10.2352/issn.2470-1173.2016.1.vda-496},
}
Figure