Distribution of Emotional Reactions to News Articles in Twitter

Distribution of Emotional Reactions to News Articles in Twitter Omar Ju´arez Gambino∗ †, Hiram Calvo†, Consuelo-Varinia Garc´ıa-Mendoza∗ ∗Instituto Polit´ecnico Nacional - ESCOM, J.D. B´atiz, Lindavista, G.A. Madero, Mexico City, 07738, Mexico †Instituto Polit´ecnico Nacional - CIC, J.D. B´atiz, Nueva Industrial Vallejo, G.A. Madero, Mexico City, 07738, Mexico b150697@sagitario.cic.ipn.mx, hcalvo@cic.ipn.mx, cvgarcia@ipn.mx Abstract Several datasets of opinions expressed by Social networks’ users have been created to explore Sentiment Analysis tasks like Sentiment Polarity and Emotion Mining. Most of these datasets are focused on the writers’ perspective, that is, the post written by a user is analyzed to determine the expressed sentiment on it. This kind of datasets do not consider the source that provokes those opinions (e.g. a previous post). In this work, we propose a dataset focused on the readers’ perspective. The developed dataset contains news articles published by three newspapers and the distribution of six predefined emotions expressed by readers of the articles in Twitter. This dataset was built aiming to explore how the six emotions are expressed by Twitter users’ after reading a news article. We show some results of a machine learning method used to predict the distribution of emotions in unseen news articles. Keywords: Twitter Sentiment Analysis, Reader’s emotions, Emotion distribution 1. Introduction Social media have fostered new ways of interaction be- tween writers and readers. The writer of a post can receive several responses from readers as soon as it is available, and then a direct interchange of opinions begins. This particu- lar form of communication has attracted attention of the Sentiment Analysis community because both writers and readers express emotions during their conversation. Twit- ter has become a popular media where users interchange opinions with strong emotional content. Since 2013 Sem- Eval workshop has organized a Twitter Sentiment Analy- sis task (Nakov et al., 2013). Specific objectives of this task have changed over time, but all released datasets since 2013 have contained tweets and annotations related to sen- timent polarity or emotions expressed on them. The Span- ish Society of Natural Language Processing (SEPLN for its acronym in Spanish) has organized a similar task but specif- ically for Spanish since 2012 (Villena Rom´an et al., 2013). The dataset for this workshop is composed by tweets and annotations corresponding to the sentiment polarity of each one. On both workshops the datasets were created focusing on writer’s perspective; this means that the proposed tasks use the text written in the opinions to predict the sentiment polarity or emotions on them. On the other hand, there have been fewer efforts focused on the reader’s perspective. In this case, the idea is to de- termine which emotions or sentiment polarity would be ex- pressed by a user as a reaction after reading a post. From this perspective, datasets comprise the written text read by users (e.g. a news article) and annotations corresponding to sentiment polarity or emotions expressed by the users. Pre- dicting the sentiment polarity or emotions of readers is con- sidered particularly difficult, because the text where …

Content truncated to 500 words