In Depth Analysis of an Online Social Network for Quitting Smoking

Social Dynamics of Substance Use in Online Social Networks for Smoking Cessation

Our multidisciplinary team of behavioral and computational social scientists will conduct an in-depth examination of a 6-year longitudinal dataset from a large online social network for smoking cessation with over 650,000 members. In addition to core components of tobacco dependence treatment, there are multiple modes for communication (e.g., forums, private messaging, wall posts), self-representation (e.g., personal profiles, blogs), and affiliation (e.g., friends). All actions are date/time stamped. The dataset contains transactional data on all users, including active events from “posters” (e.g., writing messages and blogs), passive events from “lurkers” (e.g., reading forum posts and profiles), and de-identified content of public and private communications. Embedded within this network are N=5,000 participants in an ongoing randomized trial and a distinct sample of N=1,033 participants in a 2011 cohort study. Smoking status at 3 months is known on about 60% of these participants. This unique trove of as-yet unexplored longitudinal social network data makes it possible to examine the impact of social dynamics on smoking cessation.

Using modern and innovative social computing methods, we will examine social network dynamics, sentiment, and social support in the entire network (N≈650,000). We will begin with social network analyses to construct metrics of the dynamic topology of the overall network. Next, we will conduct sophisticated sentiment analyses to construct metrics of social influence dynamics regarding “hot topics” in the online community (e.g., “cold turkey” vs. pharmacotherapy, e-cigarettes as a cessation device). We will use fine-grained text analytics to parse language related to social support (e.g., informational, appraisal) and language that signals smoking status (e.g., quit anniversaries, smoking slips). We will also examine communications related to alcohol use and problem drinking. This broad range of metrics will be used to achieve the following Specific Aims: (1) Examine social network dynamics, sentiment dynamics, and social support as predictors of 3-month smoking abstinence in the N=5,000 sample, and validate results in the N=1,033 sample; (2) Examine social network processes related to tobacco and alcohol use; and (3) Explore the correspondence between smoking status discerned from text analytics and 3-month abstinence outcomes to determine the feasibility of estimating quit rates in online social networks for cessation. This study will significantly advance methods for the study of online social networks for smoking cessation. Our work will hone computational methods that have been broadly applied in other fields but have yet to inform tobacco dependence treatment.