CECL at SemEval-2019 Task 3: Using Surface Learning for Detecting Emotion in Textual Conversations

Proceedings of the 13th International Workshop on Semantic Evaluation (SemEval-2019), pages 148–152 Minneapolis, Minnesota, USA, June 6–7, 2019. ©2019 Association for Computational Linguistics 148 CECL at SemEval-2019 Task 3: Using Surface Learning for Detecting Emotion in Textual Conversations Yves Bestgen Centre for English Corpus Linguistics Universit´e catholique de Louvain Place Cardinal Mercier, 10 1348 Louvain-la-Neuve Belgium yves.bestgen@uclouvain.be Abstract This paper describes the system developed by the Centre for English Corpus Linguistics for the SemEval-2019 Task 3: EmoContext. aimed at classifying the emotion of a user ut-It terance in a textual conversation as happy, sad, angry or other. It is based on a large number of feature types, mainly unigrams and bigrams, which were extracted by a SAS program. The usefulness of the different feature types was evaluated by means of Monte-Carlo resam- pling tests. As this system does not rest on any deep learning component, which is currently considered as the state-of-the-art approach, it can be seen as a possible point of comparison for such kind of systems. 1 Introduction This paper presents the participation of the Cen- tre for English Corpus Linguistics (CECL) in the SemEval-2019 Task 3 “EmoContext: Contextual Emotion Detection in Text”. The objective of the task is to classify the emotion of a user utterance in a textual conversation with a bot as happy, sad, angry or other. Contextual information is provided by means of the two previous utterances, the one just before, which was produced by the bot, and the first one of the triplet produced by the same person as the third. This task can be seen as a dif- ficult problem in absence of any non-written infor- mation (Gupta et al., 2017). Because of its recent and important develop- ment for the analysis of big data and especially for natural language processing, deep learning has become the preferred procedure to take up this kind of challenge (Chatterjee et al., 2019a). However, detecting emotion in texts is a well- established field of research for which unsuper- vised approaches have been proposed in content analysis (e.g. Anderson and McMaster, 1982; Bestgen, 1994) as well as less recent supervised approaches such as SVM (Chatterjee et al., 2019a; Pang and Lee, 2008). SVM has long been con- sidered as the state-of-the-art in text categorization (Joachims, 2002). Therefore, it seemed interesting to get an idea of its effectiveness for the present task in comparison to deep learning approaches that should be used by many participants in this challenge. Trying to determine what level of per- formance can be achieved with a surface learning system was thus the main focus of this study. It should be noted that the developed system does not rely on complementary training data, nor on lexical emotional norms produced manually or au- tomatically (Bestgen and Vincze, 2012), nor on se- mantic knowledge extracted from large databases or large corpora (Miller et al., 1990). However, several attempts have been made to increase its effectiveness by adding more complex features to the usual token n-grams. The …

Content truncated to 500 words