Doctoral Dissertations
Date of Award
8-2025
Degree Type
Dissertation
Degree Name
Doctor of Philosophy
Major
Data Science and Engineering
Major Professor
Seung-Hwan Lim
Committee Members
Seung-Hwan Lim, Robert Stewart, Russel Zaretzki, Audris Mockus
Abstract
Many of the systems we care about—like the economy, public health, transportation, and socioeconomics—change over time. The data that captures these changes is referred to as temporal data. The overarching goal of this work is to develop methods that make temporal data more interpretable, explorable, and actionable to analysts across many domains. Each method presented in this dissertation addresses a different modality or aspect of temporal data analysis. Together, they form a methodological toolkit that enables analysts to explore temporal patterns and extract meaningful insights from dynamic systems. Specifically, we focus on algorithmic and visualization techniques for time series and dynamic networks. We present a novel time series representation method called Shape Profiling that automatically extracts and describes time series based on intuitive shapes. The versatility of our technique allows for a wide range of query-by-content searches, such as identifying co-occurring patterns, concomitant patterns, and sequential patterns executed across various time specifications and filterings. These queries provide efficient and accurate results for practitioners that are consistent with human intuition, enabling an in-depth exploration of how shapes and trends are arranged across complex time series datasets. Our next contribution is a technique for uncovering temporal patterns in dynamic network sequences. We apply the Multidimensional Matrix Profile to snaphsot graph embeddings of dynamic networks. This data structure yields a new way to study dynamic networks based on the all-pairs similarity-join problem that has shown much success in time series data mining. Finally, we present Community Fabric, a novel visualization technique for simultaneously displaying communities and structure within dynamic networks. By hybridizing two popular network and community visualizations, Community Fabric improves upon existing state-of-the-art in several key areas. Our approach allows viewers to interpret and understand the interplay of structural change and community evolution in dynamic networks more easily. For each contribution, we apply our approaches to use cases of interest and present the results.
Recommended Citation
Ezell, Evan C., "Temporal Data Techniques for Analyst Sensemaking. " PhD diss., University of Tennessee, 2025.
https://trace.tennessee.edu/utk_graddiss/12702
Included in
Artificial Intelligence and Robotics Commons, Databases and Information Systems Commons, Data Science Commons, Graphics and Human Computer Interfaces Commons, Theory and Algorithms Commons