Choosing Your Platform for Social Media Drug Research and Improving Your Keyword Filter List

• Author: Eleanor Erin Artigiani,Nikki Adams,Eric D. Wish
• Published date: 01 July 2019
• Date: 01 July 2019
• DOI: 10.1177/0022042619833911
• Subject Matter: Articles

https://doi.org/10.1177/0022042619833911

Journal of Drug Issues

2019, Vol. 49(3) 477 –492

© The Author(s) 2019

Article reuse guidelines:

sagepub.com/journals-permissions

DOI: 10.1177/0022042619833911

journals.sagepub.com/home/jod

Article

Choosing Your Platform

for Social Media Drug

Research and Improving

Your Keyword Filter List

Nikki Adams1, Eleanor Erin Artigiani1,

and Eric D. Wish1

Abstract

Social media research often has two things in common: Twitter is the platform used and a

keyword filter list is used to extract only relevant Tweets. Here we propose that (a) alternative

platforms be considered more often when doing social media research, and (b) regardless of

platform, researchers use word embeddings as a type of synonym discovery to improve their

keyword filter list, both of which lead to more relevant data. We demonstrate the benefit of

these proposals by comparing how successful our synonym discovery method is at finding terms

for marijuana and select opioids on Twitter versus a platform that can be filtered by topic,

Reddit. We also find words that are not on the U.S. Drug Enforcement Agency (DEA) drug

slang list for that year, some of which appear on the list the subsequent year, showing that this

method could be employed to find drug terms faster than traditional means.

Keywords

social media, keyword filter list, machine learning, synonym detection

Introduction

Research in the public health domain using social media has expanded greatly in recent years. In

particular, many scholars have begun taking advantage of social media’s big data properties by

using automated and semiautomated methods to analyze these data. This research spans many

topics, such as sentiment analysis of drugs: whether people feel positively or negatively about a

particular drug (e.g., Cavazos-Rehg et al., 2015; Daniulaityte et al., 2016), whether people are

misusing prescription drugs (e.g., Chary et al., 2017), and whether some characteristics of the

drug can be correlated to its legalization status in different states (e.g., Daniulaityte et al., 2015).

The National Institute on Drug Abuse (NIDA)–funded National Drug Early Warning System

(NDEWS) sought to tap into this pool of data as well to advance its mission to monitor indicators

of emerging drug trends.

NDEWS staff, in collaboration with staff at the Center for Advanced Study of Language

(CASL) at the University of Maryland in College Park, hypothesized that an important step in

being able to monitor drug trends was to be able to keep pace with the vocabulary around drugs

1University of Maryland, College Park, USA

Corresponding Author:

Nikki Adams, Center for Advanced Study of Language, University of Maryland, 7005 52nd Ave, College Park, MD

20742, USA.

Email: nadams2@umd.edu

833911JODXXX10.1177/0022042619833911Journal of Drug IssuesAdams et al.

research-article2019

478 Journal of Drug Issues 49(3)

and drug use. They used machine learning on data from Twitter to demonstrate that new drug

terms for marijuana could be discovered using big data methods (Simpson, Adams, Brugman,

& Conners, 2018). The results, however, were less impressive for other types of drugs. The aim

of the research we present here is to improve on that performance by looking at another social

media platform, Reddit, and comparing its utility for discovering drug terms in Twitter for both

marijuana and opioids. In doing so, we hope to provide methods and recommendations that

other drug researchers can use as they do their research on social media.

Most social science research has two things in common: (a) Twitter is used as the data source,

and (b) a keyword filter list is used, which extracts only from Twitter those tweets containing a

term on that list. Against that background, our aim here is to demonstrate how to improve such

research using social media in two ways. First, we would like to bring more attention to the idea

of using other corpora (i.e., data sets of text used for language analysis), in particular, to using

Reddit. We hypothesize that Reddit would be a better platform for drug term exploration, espe-

cially for opioids, as we could focus on a drug topic subset of the platform; this hypothesis turns

out to be supported. Second, we show that, regardless of which corpus (text data set) a researcher

uses, one can improve one’s keyword filter to increase the number of relevant posts for exami-

nation by doing some synonym discovery before deciding on that final list. The ability to dis-

cover (perhaps previously unknown) synonyms for drugs has implications for anyone doing

research or working in a field that requires knowing how people refer to drugs. This latter goal

also has implications for anyone wanting to learn more about the language used around drugs

(or, in fact, any topic).

The method we use to achieve both of these goals is a machine-learning algorithm for word

embeddings, Word2Vec. In short, in this algorithm, the corpus data are used to model how

words are related to one another based on how those words are used in context. As a result, a

researcher can then take terms that he or she already knows and find out terms that are like it,

such as synonyms, some of which will be ones the researcher had not thought of in advance.

We build these word-embedding models based on Twitter and Reddit and compare them. The

results of our comparison show that we can use synonym discovery to find “new” terms, with

“new” in this context being terms not already on the U.S. Drug Enforcement Agency (DEA)

list of drug slang words (DEA Intelligence Report, 2017). Simpson et al. (2018) used such

models for uncovering marijuana-related terms on Twitter. We do it for marijuana and opioid

terms on Twitter and Reddit, and we show that this method reveals new terms on both social

media platforms.

Overall, our results show that Reddit outperforms Twitter for synonym detection for the terms

we selected. The Reddit model was built using drug-related subforums only, making the data less

noisy. Filtering by keyword sets limits on what is ultimately analyzed based on terms that are

already known. This is particularly limiting for term discovery, where the goal is finding unknown

terms. For this reason, choosing a social media platform that is organized by topic can be benefi-

cial as it can be filtered by topic while not being filtered by keyword.

Of course, researchers should still choose Twitter when the research question calls for it.

Examples of reasons for using Twitter are that the researcher wants to analyze geographic distri-

bution of information as Twitter allows geolocation in its metadata, a feature most other social

media sites do not have. Our research findings, however, show there is good reason to consider

other sources like Reddit, which also provides a large user base and which, unlike Twitter, allows

the researcher to restrict by topic without restricting by keyword.

Background

Why should researchers care to learn drug terms for social media research beyond those that they

already know? What can be gained by discovering new terms for a substance? First, as stated, if