Exploring Multivariate Event Sequences with an Interactive Similarity Builder

Abstract

Similarity‐based exploration is an effective method in knowledge discovery. Faced with multivariate event sequence data (MVES), developing a satisfactory similarity measurement for a specific question is challenging because of the heterogeneity introduced by numerous attributes with different data formats, coupled with their associations. Additionally, the absence of effective validation feedback makes judging the goodness of a measurement scheme a time‐consuming and error‐prone procedure. To free analysts from tedious programming to concentrate on the exploration of MVES data, this paper introduces an interactive similarity builder, where analysts can use visual building blocks for assembling similarity measurements in a drag‐and‐drop and incremental fashion. Based on the builder, we further propose a visual analytics framework that provides multi‐granularity visual validations for measurement schemes and supports a recursive workflow for refining the focus set. We illustrate the power of our prototype through a case study and a user study with real‐world datasets. Results suggest that the system improves the efficiency of developing similarity measurements and the usefulness of exploring MVES data.

FlexEvent: going beyond Case‐Centric Exploration and Analysis of Multivariate Event Sequences

“…Events can be filtered using regular expressions [ZDFD15, CvW17] or nodelink diagrams [KPS15]. To support additional attribute exploration and analysis, data distributions of the attributes are shown as small charts on the side [CvW17, XSZX22]. For initial data insight, we also provide users with attribute distribution visualizations using scented widgets [WHA07].…”

“…All reviewed work that integrate the multivariate data focuses on specific use cases with no more than fifteen attributes. Except for Xu et al [XSZX22] but their focus is on distance measures.…”

“…Methods that integrate the multivariate data, are the works of Wu et al [WGW*20,WLG*21], Wang et al [WMH*21], Loorak et al [LPK*15], Xu et al [XSZX22] and Bernard et al [BSKR18]. For example, they used glyphs in a table [LPK*15] or summary statistics in aggregated blocks [BSKR18].…”