EPITA-ADAPT at SemEval-2019 Task 3: Detecting emotions in textual conversations using deep learning models combination

Proceedings of the 13th International Workshop on Semantic Evaluation (SemEval-2019), pages 215–219 Minneapolis, Minnesota, USA, June 6–7, 2019. ©2019 Association for Computational Linguistics 215 EPITA-ADAPT at SemEval-2019 Task 3: Using Deep Sentiment Analysis Models and Transfer Learning for Emotion Detection in Textual Conversations Abdessalam Bouchekif 1 abdessalam.bouchekif@epita.fr Praveen Joshi 2 praveen.joshi@mycit.ie 1 LSE, EPITA Graduate School of Computer Science, France 2 ADAPT Centre, Cork Institute of Technology, Cork, Ireland 3 University Abou Bekr Belkaid-Tlemcen, Algeria Latifa Bouchekif 3 latifa.bouchekif@univ-tlemcen.dz Haithem Afli 2 haithem.afli@cit.ie Abstract Messaging platforms like WhatsApp, Fa- cebook Messenger and Twitter have gained recently much popularity owing to their ability in connecting users in real-time. The content of these textual messages can be a useful re- source for text mining to discover and unhide various aspects, including emotions. In this paper we present our submission for SemEval 2019 task ’EmoContext’. The task consists of classifying a given textual dialogue into one of four emotion classes : Angry, Happy, Sad and Others. Our proposed system is based on the combination of different deep neural networks techniques. In particular, we use Recurrent Neural Networks (LSTM, B-LSTM, GRU, B-GRU), Convolutional Neural Network (CNN) and Transfer Learning (TL) methods. Our final system, achieves an F1µ score of 74.51% on the subtask evaluation dataset. 1 Introduction The world of text conversations has undergone drastic changes during the last few years. The ge- neration and sharing of such information have be- come much easier than before with the advent of popular social media platforms such as Twit- ter, Facebook, etc. According to Statistica 1 What- sApp is the most popular mobile messaging app in the world with one billion monthly active users. Facebook Messenger closely follows with 900 million monthly active users. The content of these messages can be useful resource for text mining to discover and unhide various aspects, including emotions (Chatterjee et al., 2019a). Capturing and analysing these emotions from peoples conversations has raised growing interest within the scientific community in varied fields like cognitive and social psychology, signal pro- 1. https://www.statista.com/topics/ 1145/internet-usage-worldwide/ cessing and natural language processing (Gupta et al., 2017; Zhang et al., 2018; Majumder et al., 2018). The goal of detecting emotions in SemEval2019 task3 described in (Chatterjee et al., 2019b) is to classify a given conversation into one of four classes - happy, sad, angry and others, that best represents the mental state of the users. This can be seen as a multiclass classification problem. In this paper, we propose an approach to detect emotions like happy, sad or angry in textual mes- sages using a combination of deep learning mo- dels. We apply, also, a transfer learning approach, from a model trained on a similar task consists on the prediction of the sentiment of the conversation, i.e. positive, negative or neutral. Then, the pre- trained model is re-used to classify the dialogue into one of four classes : happy, sad, angry and others. The rest of the paper is organized as follows. Section 2 provides a brief …

Content truncated to 500 words