EmoDet at SemEval-2019 Task 3: Emotion Detection in Text using Deep Learning

Proceedings of the 13th International Workshop on Semantic Evaluation (SemEval-2019), pages 200–204 Minneapolis, Minnesota, USA, June 6–7, 2019. ©2019 Association for Computational Linguistics 200 EmoDet at SemEval-2019 Task 3: Emotion Detection in Text using Deep Learning Hani Al-Omari Malak Abdullah Nabeel Bassam alomarihani1997@gmail.com mabdullah@just.edu.jo nabeelbassam98@gmail.com Department of Computer Science Jordan University of Science and Technology Irbid, Jordan Abstract Task 3, EmoContext, in the International Workshop SemEval 2019 provides training and testing datasets for the participant teams to detect emotion classes (Happy, Sad, Angry, or Others). This paper proposes a participat- ing system (EmoDet) to detect emotions us- ing deep learning architecture. The main input to the system is a combination of Word2Vec word embeddings and a set of semantic fea- tures (e.g. from AffectiveTweets Weka- package). The proposed system (EmoDet) en- sembles a fully connected neural network ar- chitecture and LSTM neural network to ob- tain performance results that show substantial improvements (F1-Score 0.67) over the base- line model provided by Task 3 organizers (F1- score 0.58). 1 Introduction The past decades have seen an explosive growth of user-generated content through social media platforms. People are expressing online their feelings and opinions on a variety of topics on a daily basis. Tracking and analyzing public opin- ions from social media can help to predict certain political events or predicting people’s attitude towards certain products. Therefore, detecting sentiments and emotions in text have gained a considerable amount of attention(Mohammad et al., 2018). Researchers and scientists in different fields considered this a promising topic (Abdullah et al., 2018; Liu, 2012). Many machine learning approaches have been used to detect and predict emotions and sentiments. Recently, the deep neural network (DNN) is attracting more researchers as they have been benefited from the high-performance graphics processing unit (GPU) power (Abdullah et al., 2018; Dos Santos and Gatti, 2014). The shared task (Task 3: ”EmoContext”) in SemEval-2019 workshop has been designed for understanding emotions in textual conversations (Chatterjee et al., 2019). In this task, the partic- ipants are given a textual dialogue i.e. a user ut- terance along with three turns of context. The par- ticipant teams have to classify the emotion of user utterance as one of the emotion classes: Happy, Sad, Angry or Others. Further details about Task 3 and the datasets appear in Section 3. This paper describes our team approach to de- tect and classify emotions. The input has been rep- resented as word vectors (Mikolov et al., 2013b) and a set of different features which are applied to different neural network architecture to obtain the results. The performance of the system shows sub- stantial improvements F1-Score over the baseline model provided by Task 3 organizers. The remainder of this research paper is orga- nized as follows: Section 2 gives a brief overview of existing work on social media emotion and sen- timent analyses. Section 3 presents the require- ments of SemEval Task3 and examines our pro- posed system to determine the presence of emo- tion in conversational text. Section 4 …

Content truncated to 500 words