Abstract
Social media has become an integral part of modernday society. With increasingly digital societies, individuals have become more familiar and comfortable in using Online Social Networks (OSNs) for just about every aspect of their lives. This higher level of comfort leads to users spilling their emotions on OSNs and eventually their private information. In this work, we aim to investigate the relationship between users' emotions and private information in their tweets. Our research question is whether users' emotions, expressed in their tweets, affect their likelihood to reveal their own private information (privacy leakage) in subsequent tweets. In contrast to existing survey-based approaches, we use an inductive, data-driven approach to answer our research question. We use state-of-the-art techniques to classify users' emotions, and privacy scoring and employ a new technique involving BERT for binary detection of sensitive data. We use two parallel classification frameworks: one that takes the user's emotional state into account and the other for the detection of sensitive data in tweets. Consecutively, we identify individual cases of correlation between the two. We bring the two classifiers together to interpret the changes in both factors over time during a conversation between individuals. Variations were found with respect to the kinds of private information revealed in different states. Our results show that being in negative emotional states, such as sadness, anger or fear, leads to higher privacy leakage than otherwise.