GenSMT at SemEval-2019 Task 3: Contextual Emotion Detection in tweets using multi task generic approach

Proceedings of the 13th International Workshop on Semantic Evaluation (SemEval-2019), pages 225–229 Minneapolis, Minnesota, USA, June 6–7, 2019. ©2019 Association for Computational Linguistics 225 GenSMT at SemEval-2019 Task 3: Contextual Emotion Detection in tweets using multi task generic approach Dumitru C. Bogdan Faculty of Mathematics and Computer Science, University of Bucharest Human Language Technologies Research Center, University of Bucharest bogdan27182@gmail.com Abstract In this paper, we describe our participation in SemEval-2019 Task 3: EmoContext - A Shared Task on Contextual Emotion Detection in Text. We propose a three layer model with a generic, multi-purpose approach that without any task specific optimizations achieve com- petitive results (f1 score of 0.7096) in the EmoContext task. We describe our develop- ment strategy in detail along with an exposi- tion of our results. 1 Introduction In recent years, emotion detection in text has be- come more popular due to its vast potential appli- cations in marketing, artificial intelligence, educa- tion, politics, psychology, human-computer inter- action, etc. Social platforms like Twitter and Face- book, enabled access to huge amount of textual data facilitating both theoretical and experimental research. Consequently, sentiment analysis and emotion detection have gained the interest of researchers in natural language processing and other fields. While sentiment analysis refers to classifying a subjective text as positive, neutral, or negative; emotion detection identify specific types of feel- ings such as anger, joy, fear, or sadness. SemEval is the International Workshop on Se- mantic Evaluation that has evolved from SensE- val. The purpose of this workshop is to evalu- ate semantic analysis systems. Task 3 (Chatterjee et al., 2019) in this workshop, presents the task of detecting contextual emotion in a given three turn conversation where every turn is a short phrase (or tweet). Our proposed model factors a generic approach to sentiment and emotion detection in tweets thus being suitable for several tasks in 2019 SemEval competition like task 3, task 5 or task 6. By not exploiting to the fullest specificity of task 3 (be- ing a three turn conversation) and by not includ- ing transfer learning from any similar task we as- sumed a lower accuracy, being confident that in a longer term this approach will be useful for a large class of emotion detection tasks in text. We leverage the latest developments in natu- ral language processing, like deep contextualized word representation and a specially build sub- model (encoder) as we will see in section 4. 2 Previous work Sentiment classification, the task of detecting whether a text is positive or negative, has a long history of research. From the moment when (Pak and Paroubek, 2010) constructed a corpus of tweets for this task research in this area, on this type of corpus, increased considerably. Classify- ing tweets according to sentiment has many appli- cations in political science, social sciences, mar- ket research, and many others (Mart´ınez-C´amara et al., 2012). It was only natural to extend research in this area to include a more fine grained classification of text, more than positive …

Content truncated to 500 words