High-resolution temporal representations of alcohol and tobacco behaviors from social media data

Tom Huang, Anas Elghafari, Kunal Relia, Rumi Chunara

Research output: Contribution to journalArticle

Abstract

Understanding tobacco- and alcohol-related behavioral patterns is critical for uncovering risk factors and potentially designing targeted social computing intervention systems. Given that we make choices multiple times per day, hourly and daily patterns are critical for better understanding behaviors. Here, we combine natural language processing, machine learning and time series analyses to assess Twitter activity specifically related to alcohol and tobacco consumption and their sub-daily, daily and weekly cycles. Twitter self-reports of alcohol and tobacco use are compared to other data streams available at similar temporal resolution. We assess if discussion of drinking by inferred underage versus legal age people or discussion of use of different types of tobacco products can be differentiated using these temporal patterns. We find that time and frequency domain representations of behaviors on social media can provide meaningful and unique insights, and we discuss the types of behaviors for which the approach may be most useful.

Original languageEnglish (US)
Article number54
JournalProceedings of the ACM on Human-Computer Interaction
Volume1
Issue numberCSCW
DOIs
StatePublished - Nov 1 2017

    Fingerprint

Keywords

  • Alcohol
  • Behavior
  • Health
  • Natural language processing
  • Social media
  • Time-series
  • Tobacco

ASJC Scopus subject areas

  • Computer Networks and Communications
  • Human-Computer Interaction
  • Social Sciences (miscellaneous)

Cite this