Emotion analysis and detection during COVID-19

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 6938–6947 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 6938 Emotion Analysis and Detection during COVID-19 Tiberiu Sosea♣ Chau Pham♢ Alexander Tekle♡ Cornelia Caragea♣† Junyi Jessy Li♠† ♣Computer Science, University of Illinois at Chicago ♢Computer Science, Colgate University ♡Computer Science, The University of Texas at Austin ♠Linguistics, The University of Texas at Austin {tsosea2,cornelia}@uic.edu, cpham@colgate.edu, alexander.tekle@utexas.edu, jessy@austin.utexas.edu Abstract Understanding emotions that people express during large-scale crises helps inform policy makers and first responders about the emotional states of the population as well as provide emotional support to those who need such support. We present COVIDEMO, a dataset of ∼3,000 English tweets labeled with emotions and temporally distributed across 18 months. Our analyses reveal the emotional toll caused by COVID-19, and changes of the social narrative and associated emotions over time. Motivated by the time-sensitive nature of crises and the cost of large-scale annotation efforts, we examine how well large pre-trained language models generalize across domains and timeline in the task of perceived emotion prediction in the context of COVID-19. Our analyses suggest that cross-domain information transfers occur, yet there are still significant gaps. We propose semi-supervised learning as a way to bridge this gap, obtaining significantly better performance using unlabeled data from the target domain. We make our code and data available at https://github.com/tsosea2/CovidEmo. Keywords: Emotion Detection, Covid-19, Twitter 1. Introduction We live in unprecedented times caused by a coron- avirus: the COVID-19 pandemic. This pandemic has forced extremely rapid changes in our daily lives in the push to stem the spread of the COVID-19 virus. Many of us have been uprooted, disrupted and dis- tanced from family, friends and colleagues. We have transitioned in no time into a world that is suddenly more virtual than personal, sacrificing many of the daily rhythms and joys of life. These events cou- pled with the dramatic lifestyle changes consequently led to vast amounts of data generated on social me- dia platforms such as Twitter. Understanding emotions that people increasingly express on social media during large-scale crises can have wide-ranging implications, from promoting a deeper understanding of the society to informing policy makers and first responders about the emotional states of the population (Dennis et al., 2006; Fraustino et al., 2012). In Natural Language Pro- cessing, multiple datasets have been proposed to detect emotions on social media (Mohammad, 2012; Wang et al., 2012; Mohammad and Kiritchenko, 2015; Volkova and Bachrach, 2016; Abdul-Mageed and Ungar, 2017; Demszky et al., 2020), including from hurricane dis- asters (Schulz et al., 2013; Desai et al., 2020). Re- cent studies (Kabir and Madria, 2021; Imran et al., 2020; Awais et al., 2020) propose the study of emo- tions in COVID-19 and introduce various datasets an- notated with emotions. However, these datasets are not available to the large public. Moreover, in this pa- per, we study COVID-19 from a completely different standpoint: We find that the data distribution of COVID tweets …

Content truncated to 500 words