SpanEmo: Casting Multi-label Emotion Classification as Span-prediction

Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics, pages 1573–1584 April 19 - 23, 2021. ©2021 Association for Computational Linguistics 1573 SpanEmo: Casting Multi-label Emotion Classification as Span-prediction Hassan Alhuzali Sophia Ananiadou National Centre for Text Mining Department of Computer Science, The University of Manchester, United Kingdom {hassan.alhuzali@postgrad.,sophia.ananiadou@}manchester.ac.uk Abstract Emotion recognition (ER) is an important task in Natural Language Processing (NLP), due to its high impact in real-world applications from health and well-being to author profiling, consumer analysis and security. Current ap- proaches to ER, mainly classify emotions inde- pendently without considering that emotions can co-exist. Such approaches overlook poten- tial ambiguities, in which multiple emotions overlap. We propose a new model “SpanEmo” casting multi-label emotion classification as span-prediction, which can aid ER models to learn associations between labels and words in a sentence. Furthermore, we introduce a loss function focused on modelling multiple co- existing emotions in the input sentence. Exper- iments performed on the SemEval2018 multi- label emotion data over three language sets (i.e., English, Arabic and Spanish) demon- strate our method’s effectiveness. Finally, we present different analyses that illustrate the benefits of our method in terms of improving the model performance and learning meaning- ful associations between emotion classes and words in the sentence1. 1 Introduction Emotion is essential to human communication, thus emotion recognition (ER) models have a host of ap- plications from health and well-being (Alhuzali and Ananiadou, 2019; Arag´on et al., 2019; Chen et al., 2018) to consumer analysis (Alaluf and Il- louz, 2019; Herzig et al., 2016) and user profil- ing (Volkova and Bachrach, 2016; Mohammad and Kiritchenko, 2013), among others. Interest in this area has given rise to new NLP approaches aimed at emotion classification, including single-label and multi-label emotion classification. Most ex- isting approaches for multi-label emotion classi- 1Source code is available at https://github.com/ hasanhuz/SpanEmo fication (Ying et al., 2019; Baziotis et al., 2018; Yu et al., 2018; Badaro et al., 2018; Mulki et al., 2018; Mohammad et al., 2018; Yang et al., 2018) do not effectively capture emotion-specific associa- tions, which can be useful for prediction, as well as learning of association between emotion labels and words in a sentence. In addition, standard approaches in emotion classification treat individ- ual emotion independently. However, emotions are not independent; a specific emotive expression can be associated with multiple emotions. The existence of association/correlation among emo- tions has been well-studied in psychological the- ories of emotions, such as Plutchik’s wheels of emotion (Plutchik, 1984) that introduces the notion of mixed and contrastive emotions. For example, “joy” is close to “love” and “optimism”, instead of “anger” and “sadness”. # Sentence GT S1 well my day started off great the mocha machine wasn’t working @ mcdonalds. anger, dis- gust, joy, sadness S2 I’m doing all this to make sure you smiling down on me bro. joy, love, optimism Table 1: Example Tweets from SemEval-18 Task 1. GT represents the ground truth labels. Consider S1 in Table 1, which …

Content truncated to 500 words