Characterizing Exploratory Behaviors on a Personal Visualization Interface Using Interaction Logs

Poorna TalkadSukumar, Gonzalo J. Martínez, Ted Grover, Gloria Mark, Sidney K. D’Mello, Nitesh V. Chawla, Stephen M. Mattingly, Aaron Striegel2020 · DOI

Abstract

Personal visualizations present a separate class of visualizations where users interact with their own data to draw inferences about themselves. In this paper, we study how a realistic understanding of personal visualizations can be gained from analyzing user interactions. We designed an interface presenting visualizations of the personal data gathered in a prior study and logged interactions from 369 participants as they each explored their own data. We found that the participants spent different amounts of time in exploring their data and used a variety of physical devices which could have affected their engagement with the visualizations. Our findings also suggest that the participants made more comparisons between their data instances than with the provided baselines and certain interface design choices, such as the ordering of options, influenced their exploratory behaviors.

Introduction

While the visualization community has spent decades studying how analysts use "generic" data, personal visualization represents a distinct domain. When data concerns an individual's own sleep, steps, or heart rate, the motivation for exploration shifts from objective discovery to subjective self-reflection. However, capturing how people naturally explore their own data in realistic settings is challenging, as most studies are small-scale or lab-based. This research addresses the gap in understanding "in-the-wild" exploratory behavior by analyzing low-level interaction logs from a large-scale personal data dashboard.

The Tesserae Personal Data Explorer

The study leveraged data from the Tesserae Project, a year-long study of information workers. The researchers developed a web-based interface using D3.js to visualize time-series data through line graphs. The interface provided physical activity, sleep, heart-rate variability (HRV), and location-based data collected via wearables and Bluetooth beacons.

The Tesserae Personal Data Explorer interface, featuring a sidebar for navigation, a central time-series visualization, and explanatory text.

To analyze user behavior, interactions were categorized into four semantic operations:

  • Identify: Interacting with the user's own data points.
  • Inspect: Interacting with baseline data points (Cohort or "All" participants).
  • Explore: Switching between different data features (e.g., from "Sleep" to "Steps").
  • Elaborate/Abstract: Toggling between weekly and monthly summaries.

Of the 564 participants invited to view their personal dashboards, 369 (65%) accessed the interface and generated the interaction logs used for this analysis.

Key Insights and Interaction Patterns

Analysis of the logs revealed several critical behaviors regarding how non-experts engage with personal data:

Internal Comparisons vs. Baselines

Users demonstrated a significant preference for internal comparisons over social baselines. Participants were far more interested in comparing their own performance over time (e.g., today vs. yesterday) than comparing themselves to the "average" person. The logs recorded 2,343 sequences of comparing personal data points over time, compared to only 678 sequences involving baseline comparisons.

Design-Driven Exploration

Interface design choices heavily dictated the path of user exploration. The "Activity" metric was the default view and received the highest volume of interaction. Furthermore, 70% of participants explored the six available features in the exact linear order they were listed in the menu. This suggests a "path of least resistance" bias, where users follow the UI's structure rather than seeking specific, pre-planned insights.

Device and Text Engagement

Device type significantly impacted engagement. While desktop users performed more precise "identify" and "inspect" operations, mobile users had shorter sessions and fewer interactions, likely due to the difficulty of precise touch interactions on a graph designed for mouse hovers. Additionally, users largely ignored explanatory text; the median time spent in the description section was a mere 0.39 seconds, indicating that users dive straight into visual data.

Results and Engagement Variance

Exploration time was highly skewed across the participant pool. While the median session lasted 90.61 seconds, total exploration time per participant ranged from less than a second to over 24 minutes.

Distribution of total exploration time across 369 participants, showing a highly skewed engagement pattern.

Retention was also notable; 89 participants returned for more than two sessions, and 32 participants returned more than three times. Large displays with mouse interaction led to the highest engagement (median 91.07s per session), while mobile sessions were shorter (median 66.58s).

Discussion and Future Work

The findings suggest that personal visualization tools should prioritize "glanceable" designs, especially for mobile users who naturally gravitate toward these platforms. To counter the linearity bias, future interfaces might explore non-list layouts that encourage more agency in data discovery.

While this study provides a robust look at interaction logs, limitations include the relatively short 18-day logging window and the desktop-centric UI design. Future research should combine these quantitative logs with qualitative "think-aloud" sessions to understand the specific motivations behind user interactions with personal data peaks and trends.

Summary updated

Abstract

Personal visualizations present a separate class of visualizations where users interact with their own data to draw inferences about themselves. In this paper, we study how a realistic understanding of personal visualizations can be gained from analyzing user interactions. We designed an interface presenting visualizations of the personal data gathered in a prior study and logged interactions from 369 participants as they each explored their own data. We found that the participants spent different amounts of time in exploring their data and used a variety of physical devices which could have affected their engagement with the visualizations. Our findings also suggest that the participants made more comparisons between their data instances than with the provided baselines and certain interface design choices, such as the ordering of options, influenced their exploratory behaviors.
BibTeX
@misc{talkadsukumar2020characterizing,
  author = {Poorna TalkadSukumar and Gonzalo J. Martínez and Ted Grover and Gloria Mark and Sidney K. D’Mello and Nitesh V. Chawla and Stephen M. Mattingly and Aaron Striegel},
  title = {{Characterizing Exploratory Behaviors on a Personal Visualization Interface Using Interaction Logs}},
  year = {2020},
  doi = {10.31219/osf.io/axqwc},
}

Loading related content...

Figure