Interactive and Iterative Discovery of Entity Network Subgraphs

TL;DR

This paper introduces an interactive framework that combines a maximum entropy model with visualization to enable users to discover surprising subgraph patterns in entity graphs, facilitating human-in-the-loop data exploration.

Contribution

It proposes a novel maximum entropy model for subjective subgraph discovery and integrates it into an interactive visualization system for explainable graph analysis.

Findings

Faster discovery of interesting subgraphs through user interaction.

Enhanced explainability of subgraph patterns via model-guided exploration.

Effective application demonstrated on real datasets.

Abstract

Graph mining to extract interesting components has been studied in various guises, e.g., communities, dense subgraphs, cliques. However, most existing works are based on notions of frequency and connectivity and do not capture subjective interestingness from a user's viewpoint. Furthermore, existing approaches to mine graphs are not interactive and cannot incorporate user feedbacks in any natural manner. In this paper, we address these gaps by proposing a graph maximum entropy model to discover surprising connected subgraph patterns from entity graphs. This model is embedded in an interactive visualization framework to enable human-in-the-loop, model-guided data exploration. Using case studies on real datasets, we demonstrate how interactions between users and the maximum entropy model lead to faster and explainable conclusions.