Emotion -and Area- Driven Topic Shift Analysis in Social Media Discussions Kamil Topal Electrical Enginnering and Computer Science Case Western Reserve University Cleveland, OH Email: kxt147@case.edu Mehmet Koyuturk Electrical Enginnering and Computer Science Case Western Reserve University Cleveland, OH Email: mxk331@case.edu Gultekin Ozsoyoglu Electrical Enginnering and Computer Science Case Western Reserve University Cleveland, OH Email:tekin@case.edu 2016 IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining (ASONAM) IEEE/ACM ASONAM 2016, August 18-21, 2016, San Francisco, CA, USA 978-1-5090-2846-7/16/$31.00 c  2016 IEEE Abstract—Internet-based social media platforms allow indi- viduals to discuss/comment on the ”topic” of an article in an interactive manner. The topic of a comment/reply in these dis- cussions occasionally shifts, sometimes drastically and abruptly, other times slightly, away from the topic of the article. In this paper we study the phenomena of topic shifts in article- originated social media comments, and identify quantitatively the effects on topic shifts of comments (i) emotion levels (of various emotion dimensions), (ii) topic areas, and (iii) the structure of the discussion tree. We show that, with a better understanding of the topic shift phenomena in comments, automated systems can easily be built to personalize and cater to the comment-browsing and comment-viewing needs of different users. Index Terms—Social Media Analysis, Topic Shift, Emotion Analysis I. I NTRODUCTION Internet-based social media platforms allow individuals to discuss/comment on a ”topic” of their choice in an interactive manner. Usually, social media discussions are started by an article on the web that covers an event, a product, a situation, etc. Comments in these discussions have no size restrictions, allowing people to express their opinions more completely as compared to twitter tweets. While comments in a social media discussion start, usually, around the topic of an article, it is not uncommon for the discussion topic to shift during comments and replies of the ”discussion”, sometimes drastically and abruptly, other times slightly, away from the original topic. This has been a problem, and, in fact, due to large numbers of unrelated, inflammatory, or uncivilized comments, as well as the large numbers of comments on some popular articles, news websites and blogs have started to eliminate their comment sections [1]. This is an unfortunate action as readers of a discussion, not just the commenters, gather highly useful information by the simple act of reading these, sometimes informed, comments, and form more informed opinions themselvesa significant loss for both readers and those websites and blogs that eliminate comments from their software systems. We hypothesize in this paper that there are three causes for topic shifts in comments: emotion levels, the specific area of the article (e.g., sports, politics, cancer, etc.), and the structure of the ”discussion” (comment) ”tree”. These three factors collectively play a role on topic shifts in comments; and, understanding their roles in more depth can lead to building better automated comment viewing software systems that help readers sift through large numbers of comments and gather information more effectively. Motivated by our main hypothesis, we study the phenomena of topic shifts in article-originated social media comments. We attempt to identify quantitatively the effects on topic shifts of comments (i) emotion levels (of various emotion dimensions), (ii) topic areas, and (iii) the structure of the discussion tree. We show that, with a better understanding of the topic shift phenomena in comments, automated systems can easily be built to personalize and cater to the comment- browsing and comment-viewing needs of different users: users can be provided with options in real-time to (a) selectively view and reply to comments or discussion threads that are ”on the topic”, or within a range of either the original article or a specific, possibly shifted, comment of interest within the discussion tree, (b) link and view discussions of interest in temporal order even when they belong to different discussion threads within the discussion tree, (c) prune the discussion tree in real-time by specifically eliminating those discussion threads that are of no interest to them, or (d) view comments from all over the discussion tree that may have shifted from the original topic in a certain way, such as shifted to a certain ”drifted topic”. Due to space restrictions, this paper only discusses items (b) and (d). For our experimental study, we have collected about 580,000 news article comments on ten topics in different areas (though, due to space restrictions, we only discuss results of six topics), and analyzed the effects of three factors on topic shift: (i) the comments location within the discussion tree–in terms of both the level and path of the comment within the tree, (ii) comments emotion dimensions (i.e., sensitivity, aptitude, attention and pleasantness) and the associated emotion levels (e.g., for the sensitivity dimension, the six levels are rage, anger, annoyance, apprehension, fear, and terror), and (iii) the topic area (e.g., sports, politics, or health). We have found that: • In terms of a comments location in the discussion tree, the first comment of the discussion tree sets the tone for all of its descendants: if it is on the topic, usually,