#ActuallyDepressed: Characterization of Depressed Tumblr Users’ Online Behavior from Rules Generation Machine Learning Technique

31st Pacific Asia Conference on Language, Information and Computation (PACLIC 31), pages 132–139 Cebu City, Philippines, November 16-18, 2017 Copyright c⃝2017 Czarina Rae Cahutay and Aileen Joan Vicente #ActuallyDepressed: Characterization of Depressed Tumblr Users’ Online Behavior from Rules Generation Machine Learning Technique Abstract The ubiquitous data provided by social networking sites paved the way for researchers to understand netizens behavior with psychological ailments such as depression. However, most of these researches are aimed at classifying users with depression using blackbox algorithms such as SVM. This does not allow data exploration or further understanding the characteristics of depressed individuals. This research aims to characterize depressed Tumblr users online behavior from rules generation . Characterization is done by collecting affective, social, cognitive and linguistic style markers collected from the respondents posts. Rules are then generated from these features using CN2 algorithm — a rules generating machine learning technique. The rules are analyzed and are compared with observations in prior literature on depressive behavior. We observed that depressed respondents in Tumblr have more photo content in posts rather than just pure textual posts, posts are more negative, and there is an evident use of self-referencing words. These characteristics are also evident in offline behavior of depressed individuals based on prior literature Keywords: Depression, Characterization, Rules Generation, CN2 Algorithm 1. Introduction According to the World Health Organization (WHO) (2001), depression is a serious mental issue and may be the second leading cause of disease by 2020. Amy Courtney (2014) argues that blogging reduces the symptoms brought about depression. Blogging allows the public to access and comment on such work, allowing additional psychological benefits. Microblogging is an easier way of blogging that allows users to create short contents shared with an audience in real-time. This creates an easier and more convenient way of sharing content and information through the web (Nations, 2015). Contents vary from text, to visual, audio, audiovisual and even the use of links to redirect to other websites. This study aims to characterize the online behavior of depressed individuals using machine learning technique. Further, this research is guided by the following questions: RQ1. What are the characteristics of depressed individuals on Tumblr in terms of their affective, behavioral, cognitive, and linguistic style attributes? RQ2. How do these characteristics compare to offline behavior of depressed individuals in literature? Are they any different? Czarina Rae C. Cahutay Aileen Joan O. Vicente University of the Philippines Cebu University of the Philippines Cebu Gorordo Avenue, Lahug, Cebu City Gorordo Avenue, Lahug, Cebu City czarinacahutay@gmail.com aovicente@up.edu.ph 132 This study focuses on Tumblr because there have not been a lot of studies focusing on adolescents social media postings. Other studies on depression online are focused on an adult age group. Moreover, we find that because Tumblr allows anonymity of users there might be a more genuine response with regards to their posting and the results that we gather since they are not bound to their true identity in person similar to the study of Warner et al (2016). …

Content truncated to 500 words