Quantifying daily rhythms with non-negative matrix factorization applied to mobile phone data.

Area of Science:

• Chronobiology
• Computational Social Science
• Human Behavior Analysis

Background:

• Human activities exhibit daily, weekly, and seasonal rhythms influenced by physiology and social factors.
• Circadian rhythms, near 24-hour biological cycles, vary in phase among individuals.
• Current chronotype categorization relies on questionnaires or manual feature extraction from activity data.

Purpose of the Study:

• To develop a novel, data-driven method for characterizing individual activity rhythms.
• To move beyond traditional chronotype classifications based solely on sleep timing.
• To analyze temporal patterns in human behavior using readily available smartphone data.

Main Methods:

• Non-negative matrix factorization (NMF) applied to time-stamped smartphone screen usage logs.
• Decomposition of activity data into emergent temporal components without prior assumptions.
• Analysis of a dataset comprising one year of mobile phone usage from 400 university students.

Main Results:

• Identification of four prominent temporal activity components: morning, night, evening, and noon.
• Demonstration that individual behavior can be represented as weights across these components.
• Finding that individuals fall on a continuous spectrum of activity timing, not discrete categories.
• Strong correlation between high morning/night component weights and actual sleep/wake times.

Conclusions:

• The proposed NMF method offers a context-dependent, data-driven approach to understanding human activity rhythms.
• Characterizing individuals based on their full daily and weekly activity patterns provides a more nuanced view than traditional chronotypes.
• Smartphone data can be effectively utilized to uncover complex behavioral temporalities.