There is lot of research activity in the area of opinion mining and sentiment analysis, which deals with the computational treatment of opinion, sentiment, and subjectivity in text. Social media websites have become increasingly popular for discussing uncomfortable topics. However, there are limited resources for mining and automatically labeling posts discussing self-disclosure. There is great incentive for a system which can be useful for monitoring emotional state of users, both for the research community as well as for mental health and business purposes.
This paper presents a case where we leverage information from psycho-lingual and crowd-sourced dictionaries to create a system which can automatically predict anonymous posts about taboo topics on a social media site (Facebook Confessions). We achieve more than 80% accuracy for the most popular taboo topics, and an overall accuracy of 61.25 % across all taboo categories. We evaluate our system in two ways: a) comparing against human-annotated posts on another anonymous social media platform YikYak b) an evaluation against existing state-of-the-art models.