National substance use patterns on Twitter.

ABSTRACT:

Purpose

We examined openly shared substance-related tweets to estimate prevalent sentiment around substance use and identify popular substance use activities. Additionally, we investigated associations between substance-related tweets and business characteristics and demographics at the zip code level.

Methods

A total of 79,848,992 tweets were collected from 48 states in the continental United States from April 2015-March 2016 through the Twitter API, of which 688,757 were identified as being related to substance use. We implemented a machine learning algorithm (maximum entropy text classifier) to estimate sentiment score for each tweet. Zip code level summaries of substance use tweets were created and merged with the 2013 Zip Code Business Patterns and 2010 US Census Data.<

SUBMITTER: Meng HW 

PROVIDER: S-EPMC5673183 | biostudies-literature | 2017

REPOSITORIES: biostudies-literature