This visual analytics workflow analyses multivariate temporal data partitioned into episodes to understand complex dynamic phenomena through progressive abstraction. The approach has three main analytical phases:
enumerate Single-attribute pattern extraction: Transform temporal sequences of attribute values within episodes into symbolic representations (using methods like SAX encoding) that capture variation patterns. Analysts iteratively refine encoding parameters (segment counts, discretization breaks) to achieve interpretable patterns. Multi-attribute pattern discovery: Apply topic modeling to collections of symbolic patterns to discover "topics" - combinations of single-attribute patterns that frequently co-occur across episodes. Analysts experiment with parameters (number of topics), assess topic interpretability, and may merge semantically similar topics. Distribution pattern analysis: Visualize and explore how multi-attribute patterns (topics) are distributed across episodes in relevant contexts (time, space, external conditions). Analysts identify higher-level distribution patterns and formulate domain insights.
This page shows an experimental modular rendering of the ATWL representation, using a proposed module … end module construct that groups related artifacts and transforms behind a named, collapsible interface (declared input/output artifacts). This construct is not yet part of the canonical ATWL infrastructure: the graphical viewer, structural viewer, and syntax checker do not recognise it, so those views are omitted here. See the original page for the canonical representation.