Social Media Analytics of Smoking Cessation Intervention: User Behavior Analysis, Classification, and Prediction

Tobacco use causes a large number of diseases and deaths in the United States. Traditional intervention programs are based on face-to-face consulting, and social support is offered to help smoking quitters control stress and achieve better intervention outcomes. However, the scalability of these traditional intervention programs is limited by time and location. With the development of Web 2.0, many intervention programs of smoking cessation are developed online to reach a wider population. QuitNet is a popular website for smoking cessation that provides different services to help users quit smoking. It builds communities on different social media for people to discuss issues of smoking cessation and provide social support for each other. In this dissertation, we develop a comprehensive study to understand user behavior and their discussion interactions in online communities of smoking cessation. We compare user features and behaviors on different social media channels, analyze user interactions from the perspective of social support exchange, and apply data mining techniques to analyze discussion content and recommend threads for users.;Health communities are developed on different types of social media. For example, QuitNet has Web forums on its own Web site while it also has its appearance on Facebook. The user participation may vary on different social media platforms. Users may also behave differently depending on the functions and design of the social media platforms. So, as the first step in this dissertation, we carry out a preliminary study to compare smoking cessation communities on different social media channels. We analyze user characteristics and behaviors in QuitNet Forum and QuitNet Facebook with statistical analysis and social network analysis. It is found that most users of QuitNet Forum are early smoking quitters, and they participate in discussions more actively than users of QuitNet Facebook. However, users of QuitNet Facebook have a wider spectrum of quitting statuses and interaction behaviors.;Second, we are interested in user behaviors and how they exchange social support in online communities. Social support is "an exchange of resources between two individuals perceived by the provider or the recipient to be intended to enhance the well-being of the recipient". As QuitNet Forum attracts much more active users than QuitNet Facebook, it provides a better platform for our research purpose. So, we focus on QuitNet Forum, developing a classification scheme through qualitative analysis to categorize discussion topics and types of social support on the forum. Patterns of user behaviors are defined and identified. Social networks are built to analyze user interactions of social support exchange. It is found that users at different quit stages have different behaviors to exchange social support, and different types of social support flow between users at different quit stages.;Discussion topics, user behaviors and patterns of social support exchanges are thoroughly analyzed. However, due to a huge amount of information on QuitNet Forum, it is difficult for users to find proper topics or peers to discuss or interact with. It would be helpful if we could apply machine learning techniques to understand user generated information in online health communities, and recommend discussion topics to users to participate in. We develop classifiers to categorize posts and comments on QuitNet Forum in terms of user intentions and social support types. User behaviors and patterns are used to help developing various feature sets. Then, we develop recommendation techniques to recommend threads for users to participate in. Based on traditional Collaborative Filtering and content-based approaches, we integrate classification results and user quit stages to develop recommendation systems. The experiments show that integrating classification results or user health statuses can achieve the best recommendation results with different percentages of unknown data.;In this dissertation, we implement all-sided studies for online smoking cessation communities, including comprehensive analytics and applications. The proposed frameworks and approaches could be applied to other health communities. In the future, we will apply more analytics and techniques to a larger data set, and develop user-end applications to serve and improve online health intervention programs and communities.