Building Deep Data through Screens: Princeton-SMART for High-Frequency Mobile Communication Research

This paper introduces the Princeton-Screenlake Mobile Analytics and Recording Tool (Princeton-SMART), an open-source mobile sensing system designed to generate deep data on communication behavior. Princeton-SMART enables unobtrusive collection of smartphone screenshots at short intervals in natural settings. We describe a research pipeline for secure data collection, preprocessing, and analysis, drawing on over 1.5 million screenshots contributed by U.S. job seekers at 3-second intervals over six weeks. We demonstrate how job-search activities can be identified using vision-language AI models and how job-search patterns vary by socioeconomic status and digital literacy. We conclude by discussing opportunities and practical considerations for integrating Princeton-SMART into mobile communication research.