CausalFlow: Visual Analytics of Causality in Event Sequences
This work addresses the need for clearer insights in domains like application logs and sports records, offering an incremental improvement over existing co-occurrence methods.
The paper tackles the problem of understanding event relations by using causation instead of co-occurrence, presenting a visualization approach called CausalFlow that integrates automatic causal discovery and a novel visualization to help users identify causal pathways in event sequences.
Understanding the relation of events plays an important role in different domains, such as identifying the reasons for users' certain actions from application logs as well as explaining sports players' behaviors according to historical records. Co-occurrence has been widely used to characterize the relation of events. However, insights provided by the co-occurrence relation are vague, which leads to difficulties in addressing domain problems. In this paper, we use causation to model the relation of events and present a visualization approach for conducting the causation analysis of event sequences. We integrate automatic causal discovery methods into the approach and propose a model for detecting event causalities. Considering the interpretability, we design a novel visualization named causal flow to integrate the detected causality into timeline-based event sequence visualizations. With this design, users can understand the occurrence of certain events and identify the causal pathways. We further implement an interactive system to help users comprehensively analyze event sequences. Two case studies are provided to evaluate the usability of the approach.