Modeling Self-Disclosure in Social Networking Sites: Understanding the Causes and Consequences of Self-Disclosure through Automatic Language Analysis

Social networking sites (SNSs) offer users a unified platform to build and maintain social connections. Quality user experience relies on understanding when people feel comfortable sharing information about themselves on SNSs because self-disclosure helps to maintain friendships, increases relationship closeness, and benefits the discloser’s health and well-being. This thesis develops a new machine learning model to measure self-disclosure in SNS communication at scale to better understand the contexts in which users of Facebook, the world’s largest social networking site, disclose a higher or lower level of personal information about themselves. The machine learning model was built using four key features, including emotional valence, social distance between the poster and people mentioned in the post, similarity of language in the post to what others are discussing, and post topics. The model performs moderately and in line with the judgments of trained coders (r=.60). I then apply the model automatically to de-identified, aggregated samples of Facebook users’ status updates and examine factors at three levels that might influence their self-disclosure: their stable, personal characteristics, the structure of their Facebook networks, and events in their lives. 

Results from this study confirm and extend earlier psychological research on the conditions associated with self-disclosure. Specifically, this study shows that women selfdisclose more than men, and users who score higher on an Impression Management scale, indicating a stronger desire to manage the impressions others have of them, self-disclose less. At the level of audience structures, results indicate that social network size negatively correlates with self-disclosure, while network density and average tie strength with friends positively correlate. However, the analysis results of product feature tests designed to make users more aware of the audience’s existence were ambiguous. 

Longitudinal analyses examining self-disclosure among Facebook users who experienced major life events indicate that positive events increase self-disclosure, whereas negative events constrain disclosure. In particular, users self-disclosed more during periods when they were experiencing the start of a new romantic relationship and self-disclosed less when experiencing a break-up. In addition, students disclosed more about themselves at 4 the start of their academic term; this peak was larger for college freshman than for college sophomores. Further, increased self-disclosure correlates with a smaller increase in users’ friend count, which indicates potential tension between audience size and disclosure. 

This thesis has both theoretical and practical implications. Theoretically, it advances our understanding of the conditions associated with variation in online self-disclosure. Practically, it provides methods for measuring self-disclosure at scale in social networking sites and guidance for SNS designers to improve their services by providing better affordances to users.