EMA at SemEval-2018 Task 1: Emotion Mining for Arabic

Proceedings of the 12th International Workshop on Semantic Evaluation (SemEval-2018), pages 236–244 New Orleans, Louisiana, June 5–6, 2018. ©2018 Association for Computational Linguistics EMA at SemEval-2018 Task 1: Emotion Mining for Arabic Gilbert Badaro, Obeida El Jundi, Alaa Khaddaj, Alaa Maarouf, Raslan Kain Hazem Hajj, Wassim El-Hajj† Department of Electrical and Computer Engineering, American University of Beirut † Department of Computer Science, American University of Beirut Beirut, Lebanon {ggb05;oae15;awk11;aim20;rhk44;hh63;we07}@aub.edu.lb Abstract While significant progress has been achieved for Opinion Mining in Arabic (OMA), very limited efforts have been put towards the task of Emotion mining in Arabic. In fact, busi- nesses are interested in learning a fine-grained representation of how users are feeling to- wards their products or services. In this work, we describe the methods used by the team Emotion Mining in Arabic (EMA), as part of the SemEval-2018 Task 1 for Affect Mining for Arabic tweets. EMA participated in all 5 subtasks. For the five tasks, several prepro- cessing steps were evaluated and eventually the best system included diacritics removal, elongation adjustment, replacement of emojis by the corresponding Arabic word, character normalization and light stemming. Moreover, several features were evaluated along with dif- ferent classification and regression techniques. For the 5 subtasks, word embeddings feature turned out to perform best along with Ensem- ble technique. EMA achieved the 1st place in subtask 5, and 3rd place in subtasks 1 and 3. 1 Introduction Emotion recognition has captured the interest of researchers for many years. Different models have been used to detect people’s emotions such as human computer interaction (HCI) (Hibbeln et al., 2017; Patwardhan and Knapp, 2017; Con- stantine et al., 2016) and their facial expressions (Trad et al., 2012; Wegrzyn et al., 2017). Re- cently, with Web 2.0, the size of textual data charged with opinions and emotions on the web has tremendously increased. Thus, researchers have been looking at automatically performing sentiment and emotion analysis from textual data. In fact, learning emotions of users is critical for different applications such as shaping mar- keting strategies (Bougie et al., 2003), providing customers with better personalized recommenda- tions for advertisements and products (Moham- mad and Yang, 2011), improving recommendation of typical recommender systems (Badaro et al., 2013, 2014c,d), tracking emotions of users to- wards politicians, movies, music, products, etc, (Pang et al., 2008), or accurately predicting stock market prices (Bollen et al., 2011). Some efforts have already been placed in de- veloping emotion classification models from text (Shaheen et al., 2014; Houjeij et al., 2012; Abdul- Mageed and Ungar, 2017). Since sentiment lex- icons helped in improving the accuracy of senti- ment classification models (Liu and Zhang, 2012; Taboada et al., 2011), several researchers are working on developing emotion lexicons for dif- ferent languages such as English, French, Chinese (Mohammad, 2017; Bandhakavi et al., 2017; Yang et al., 2007; Poria et al., 2012; Das et al., 2012; Mohammad et al., 2013; Abdaoui et al., 2017; Staiano and Guerini, 2014; Badaro et al., 2018a). There were also couple of attempts …

Content truncated to 500 words