Aluno: Jonathas Gabriel Dipp Harb
Orientadora: Profª Drª Karin Becker
Título: Using a convolutional neural network to compare emotional reactions on Twitter to mass violent events
Linha de Pesquisa: Mineração, Integração e Análise de Dados
Local: Prédio 43424 – Auditório Prof. Castilho do Instituto de Informática da UFRGS
Prof. Dr. Duncan Dubugras Alcoba Ruiz (PUCRS)
Profª. Drª. Renata de Matos Galante (UFRGS)
Profª. Drª. Viviane Pereira Moreira (UFRGS)
Presidente da Banca: Profª Drª Karin Becker
Abstract: The number of mass violent attacks, particularly mass shooting and terrorism events, has increased in recent years. Understanding the emotional reaction of the population is very important to help them to cope with the constant sense of threat and fear effectively. In this paper, we apply deep learning techniques to classify emotions expressed by Twitter users, and develop a comparative analysis of emotional reactions to twelve mass violent events, detailed using demographics information extracted from the users proﬁles. To classify the emotions, we trained a Convolutional Neural Network combining sets of automatically ﬁltered training seeds and pre-trained word embeddings. Our study compared four terrorism events and eight mass shooting incidents in terms of emotional shift and prevalent emotions; inﬂuence on emotions of gender, age, closeness to the event and number/type of victims; and terms used to express reactions. We observed similar patterns for both kinds of events, mainly in terms of prevalent emotions (anger, fear, and sadness, respectively) and inﬂuence of genderon emotions (e.g. fear for women, and anger formen). The proximity to the events is inﬂuential only in mass shooting events. Tweeters expressing fear and sadness tend to share words of empathy and support, while those expressing anger tend to use intense words of hate, intolerance and call for justice.
Keywords: Deep Learning. Convolutional Neural Network. Sentiment Analysis. Twitter. Mass Violent Events.