Event sequence data is critical across domains like healthcare and industrial operations, yet the field of event sequence visual analytics remains highly fragmented. The current landscape is rich in domain-specific solutions but poor in generalizable principles. This dissertation addresses this fragmentation by establishing methodologies to enhance generalizability across three fundamental dimensions: theoretical, computational, and empirical.

First, we introduce a theoretical, domain-agnostic task framework that provides a shared analytical vocabulary to seamlessly transfer knowledge across disparate application areas. Second, we present a computational causal framework that extends treatment effect estimation to the temporal point process setting, enabling the robust extraction of causal relations under non-stationary conditions. Finally, we provide two empirical contributions to address evaluation bottlenecks. First, we conduct a crowdsourced experiment that identifies visual complexity as a primary factor affecting human comprehension. Second, we introduce ProcVQA, a novel benchmark designed to evaluate the baseline structural comprehension of Vision-Language Models (VLMs), providing the foundation required to utilize AI as a scalable, automated evaluation tool.