Mining Patterns of Substance by Young Adults with Social Media Data

University Of North Carolina Charlotte, Charlotte NC

Investigators

Abstract

DESCRIPTION (provided by applicant): Substance abuse is a serious health issue facing alarming numbers of young adults (aged from 18 to 25), who often suffer considerable consequences (e.g., blackout, rape, HIV-related sexual risk-taking, academic failure, mental issues, and violence) as a result. To understand the trends in substance use, both national and regional surveys (such as the National Survey on Drug Use and Health) are given every year to a large number of randomly-selected individuals of different ages. Various statistics (e.g., percentage of substance use) are obtained from the survey data to determine trends of substance use. However, these traditional survey-based approaches to understanding these trends are slow, costly and inadequate. For example, the use of molly and alcopop drinks is recognized as an emerging issue, but national surveys do not capture current information about their use. This project is to explore Twitter data stream to extract public dialogues related to substance use and uncover patterns of substance use in a timely, economical and in-depth way. Due to the noisy and massive nature of Twitter data, we will develop advanced data mining methods to extract tweets related to substance use via leveraging substance slang. With these captured signals related to substance use, we will then discover various patterns of substance use, such as emerging types of substances, new combinations of substance use, and dynamic geographic and temporal distributions of substance use, among young adults. And these patterns could be discovered before the national or regional survey, so they could be used for generating early warnings of substance use. We will further explore in-depth information about the identified emerging patterns of substance use via conducting focus groups, which can be used as a basis for developing more timely interventions. In addition, we will compare and validate our results (such as the temporal and spatial distribution of substance use) from social media data with results from traditional surveys. Our innovative approach to understanding trends in substance use represents a shift from the current survey paradigm through a deep analysis of social media data. We expect that we will obtain much deeper, finer and earlier findings, and as well as consistent results with those of existing reports and survey data.