SWAP at SemEval-2019 Task 3: Emotion detection in conversations through Tweets, CNN and LSTM deep neural networks

Proceedings of the 13th International Workshop on Semantic Evaluation (SemEval-2019), pages 324–329 Minneapolis, Minnesota, USA, June 6–7, 2019. ©2019 Association for Computational Linguistics 324 SWAP at SemEval-2019 Task 3: Emotion detection in conversations through Tweets, CNN and LSTM deep neural networks Marco Polignano University of Bari A. MORO Dept. Computer Science E.Orabona 4, Italy marco.polignano@uniba.it Marco de Gemmis University of Bari A. MORO Dept. Computer Science E.Orabona 4, Italy marco.degemmis@uniba.it Giovanni Semeraro University of Bari A. MORO Dept. Computer Science E.Orabona 4, Italy giovanni.semeraro@uniba.it Abstract Emotion detection from user-generated con- tents is growing in importance in the area of natural language processing. The approach we proposed for the EmoContext task is based on the combination of a CNN and an LSTM using a concatenation of word embeddings. A stack of convolutional neural networks (CNN) is used for capturing the hierarchical hidden rela- tions among embedding features. Meanwhile, a long short-term memory network (LSTM) is used for capturing information shared among words of the sentence. Each conversation has been formalized as a list of word embed- dings, in particular during experimental runs pre-trained Glove and Google word embed- dings have been evaluated. Surface lexical features have been also considered, but they have been demonstrated to be not usefully for the classification in this specific task. The fi- nal system configuration achieved a micro F1 score of 0.7089. The python code of the sys- tem is fully available at https://github. com/marcopoli/EmoContext2019. 1 Introduction The task of emotion detection from a text is grow- ing in importance as a consequence of a large number of possible applications in personalized systems. This task can be considered as part of the sentiment analysis process also if it differs about the information collected. Sentiment Anal- ysis aims to detect the polarity (positive, nega- tive or neutral) about a topic of discussion or a specific aspect. On the contrary, Emotion Detec- tion aims to associate an emotional label to textual content to explicitly understand what is the emo- tional state of the user while writing it. The fi- nal user behaviors are strongly influenced by the emotional state which she is in. Following the studies of Ekman (Ekman et al., 1987), Plutchik (Plutchik, 1990), Parrot (Parrott and Sabini, 1990), and Frijda (Frijda and Mesquita, 1994) some emo- tions can be considered ”basics” and consequently more important than others during everyday deci- sions. Their identification is, therefore, one cru- cial aspect for applications in commerce, pub- lic health, disaster management, and trend anal- ysis (consumer understanding). In the research area of emotion detection and sentiment analy- sis, many challenges are organized every for over- coming the state-of-the-art results. SemEval 1 is one of the most famous among them and it pro- vides a large amount of data every year useful for supporting the research about the topic and com- monly considered as state-of-the-art. Recently the best results are obtained by machine learning ap- proaches (Colneriˆc and Demsar, 2018) based on recurrent neural networks (long short-term mem- ory network) …

Content truncated to 500 words