TimePool: Visually Answer "Which and When" Questions On Univariate Time Series

Tinghao Feng, Yueqi Hu, Jing Yang, Tom Polk, Ye Zhao, Shixia Liu, Zhaocong Yang

TL;DR

TimePool is a visualization prototype designed to solve the "spaghetti chart" problem in large-scale time series data. By combining traditional line charts with interactive, juxtaposed timelines, it allows users to quickly identify which entities meet specific criteria (like being in the top 10 or exceeding a threshold) and when those conditions occurred.

What Problem Does This Paper Address?

Analyzing univariate time series data—data where a single variable is tracked over time for many different entities—is a common task in fields ranging from public health to finance. However, as the number of entities grows (e.g., tracking life expectancy for 200 countries over 100 years), standard line charts become cluttered and unreadable.

The authors identify a specific gap: it is difficult for analysts to answer "Which and When" questions. For example, "Which countries were in the top 10 for life expectancy in the 1960s, and when did they drop out?" In a standard line chart, these lines overlap and cross constantly, making it nearly impossible to track individual trajectories or identify transient patterns.

Methods

TimePool introduces a multi-view interface consisting of four main components:

  1. Line Chart (Top Left): Displays the raw data. Crucially, users can directly manipulate thresholds by dragging lines or curves across this chart to define queries.
  2. Detail View (Bottom Left): A set of juxtaposed timelines (one row per entity). When an entity meets the query criteria, its segment in the timeline is colored (e.g., green for "high," red for "low"); otherwise, it remains gray.
  3. Overview (Bottom Right): A zoomed-out version of the detail view that allows for the identification of global patterns across the entire dataset.
  4. Control Panel (Top Right): For fine-tuning query parameters.

Key Innovations: * Dynamic Queries on Derived Values: Beyond raw values, TimePool supports queries based on ranking, percentage change, net change, and variance. * Sorting and Grouping: Users can sort timelines by the total duration a condition was met or group them by categories (like continents) to reveal regional trends. * Ego Mode: A comparison tool where one entity (the "ego") is used as a dynamic baseline for all others.

Datasets

The researchers evaluated TimePool using three primary datasets:

  • World Life Expectancy (WLE): Data for 200 countries from 1900–2012, sourced from Gapminder.
  • World GDP: Data for 197 countries from 1990–2012.
  • Beijing Air Quality: A case study dataset recording PM 2.5 pollutant levels.

Key Insights and Ideas

  • The Power of Juxtaposition: By moving from an "overlay" (lines on top of each other) to a "juxtaposed" (side-by-side timelines) view, the "which" and "when" become immediately visible as colored blocks, eliminating visual clutter.
  • Direct Manipulation: Allowing users to drag threshold curves directly on the line chart provides an "exploratory loop." As you move the threshold, the colored blocks in the timelines shift in real-time, helping users find meaningful cut-off points.
  • Temporal Sorting: Sorting by the length of the colored segments is a simple but effective way to distinguish between entities that are "consistently" in a state versus those that are "transiently" in that state.
  • Regional Patterns: Grouping by category (e.g., continent) immediately highlighted that while most of the world improved in life expectancy, Sub-Saharan Africa faced unique, recurring challenges that caused many countries to fall back below the 50-year threshold in the 1990s.

Key Results

The authors conducted a user study comparing TimePool to a baseline system (a standard line chart with query tools).

  • Speed: For "Extreme" tasks (finding top members), TimePool users were nearly 3x faster (43 seconds vs. 124 seconds).
  • Accuracy: TimePool significantly boosted accuracy. For comparison tasks, TimePool users achieved 96% accuracy, compared to only 63% for the baseline.
  • Expert Consensus: 18 domain experts rated the system highly (averaging ~8.9/10 for usefulness and intuitiveness), noting that it "provides an intuitive and efficient way to learn the pattern hidden in a large number of complex time series datasets."

Limitations and Future Work

  • Learning Curve: Some experts noted a slight initial "learning burden" to understand how the different views coordinate.
  • Scalability: While better than line charts, the system's vertical scrolling might become a bottleneck for datasets with thousands or millions of entities.
  • Univariate Focus: The current prototype only handles one variable at a time. Experts expressed a strong desire to extend this "Which and When" logic to multivariate data (e.g., "Which countries had high GDP and low infant mortality?").
  • Data Integration: Future versions could benefit from supporting more diverse data sources, such as NoSQL databases, and adding automated annotation features.

Summary updated

Abstract

When exploring time series datasets, analysts often pose “which and when” questions. For example, with world life expectancy data over one hundred years, they may inquire about the top 10 countries in life expectancy and the time period when they achieved this status, or which countries have had longer life expectancy than Ireland and when. This paper proposes TimePool, a new visualization prototype, to address this need for univariate time series analysis. It allows users to construct interactive “which and when” queries and visually explore the results for insights. TimePool has been evaluated through example scenarios, a formal user study, and interviews with domain experts.
BibTeX
@inproceedings{feng2023timepool,
  author = {Tinghao Feng and Yueqi Hu and Jing Yang and Tom Polk and Ye Zhao and Shixia Liu and Zhaocong Yang},
  title = {{TimePool: Visually Answer "Which and When" Questions On Univariate Time Series}},
  booktitle = {IEEE VIS (Short Papers)},
  year = {2023},
  url = {https://openreview.net/forum?id=t13yqoaJQU},
}

Loading related content...

Figure