Beyond the Mirror: Personal Analytics through Visual Juxtaposition with Other People's Data

TL;DR

The paper addresses how personal analytics can be enriched by placing an individual's calendar data in a comparative context with others. It introduces CalTrend, a visual analytics system that uses deidentified online schedule logs, life-mode labeling, and ML-driven insights (via t-SNE projections and topic modeling) to contrast individuals and groups. Through a qualitative study with two domain experts, the authors show that domain-specific mental models shape interpretations and enable identification of meaningful groups or anomalies, suggesting practical utility for cross-domain personal analytics and targeted insights. Privacy considerations and limited generalizability are acknowledged, with future work pointing to broader datasets and additional domains to strengthen generalizability and impact.

Abstract

An individual's data can reveal facets of behavior and identity, but its interpretation is context dependent. We can easily identify various self-tracking applications that help people reflect on their lives. However, self-tracking confined to one person's data source may fall short in terms of objectiveness, and insights coming from various perspectives. To address this, we examine how those interpretations about a person's data can be augmented when the data are juxtaposed with that of others using anonymized online calendar logs from a schedule management app. We develop CALTREND, a visual analytics system that compares an individuals anonymized online schedule logs with using those from other people. Using CALTREND as a probe, we conduct a study with two domain experts, one in information technology and one in Korean herbal medicine. We report our observations on how comparative views help enrich the characterization of an individual based on the experts' comments. We find that juxtaposing personal data with others' can potentially lead to diverse interpretations of one dataset shaped by domain-specific mental models.

Figures (2)

• Figure 1: The overall workflow of CalTrend.CalTrend is developed in three stages: first, we deidentify information from the schedule logs. Second, we preprocess the data by annotating life modes and defining features reflecting each user's scheduling patterns. Finally, we apply various machine learning (ML) techniques to generate insights, which are integrated into the CalTrend visual analytics system.
• Figure 2: The CalTrend interface.CalTrend is a visual analytics system designed to reveal temporal and contextual trends in online calendar logs. (A) A scatterplot presents a two-dimensional projection of calendar users, where zooming transforms dots into glyphs encoding schedule frequency, life modes, and total number of events. Hovering over a dot displays its associated textual data. From (A) and (1), the user can select a user group(or individual) of their interest (see § \ref{['subsec:interface-explain']}). (B) The analyst can adjust feature weights to modify the t-SNE projection, with black bars in the histogram indicating the user’s values for each quantified feature. (C) An hour-by-day heatmap captures daily scheduling patterns, while (D) a line graph visualizes the cumulative frequency of schedules. (E) Two topic-model-based wordclouds highlight key textual themes, and (F) a novel hour-by-week heatmap illustrates repetitive scheduling patterns, offering interactions such as keyword distribution analysis (see § \ref{['subsec:detailed_view']} for more details). Overall, these integrated views provide a comprehensive perspective on user behaviors, supporting rich domain-specific insights.