Unsupervised Extractive Summarization of Emotion Triggers

Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 9550–9569 July 9-14, 2023 ©2023 Association for Computational Linguistics Unsupervised Extractive Summarization of Emotion Triggers Tiberiu Sosea∗1 Hongli Zhan∗2 Junyi Jessy Li2 Cornelia Caragea1 1Department of Computer Science, University of Illinois Chicago 2Department of Linguistics, The University of Texas at Austin {tsosea2,cornelia}@uic.edu {honglizhan,jessy}@utexas.edu Abstract Understanding what leads to emotions during large-scale crises is important as it can provide groundings for expressed emotions and sub- sequently improve the understanding of ongo- ing disasters. Recent approaches (Zhan et al., 2022) trained supervised models to both de- tect emotions and explain emotion triggers (events and appraisals) via abstractive summa- rization. However, obtaining timely and quali- tative abstractive summaries is expensive and extremely time-consuming, requiring highly- trained expert annotators. In time-sensitive, high-stake contexts, this can block necessary responses. We instead pursue unsupervised sys- tems that extract triggers from text. First, we introduce COVIDET-EXT, augmenting (Zhan et al., 2022)’s abstractive dataset (in the con- text of the COVID-19 crisis) with extractive triggers. Second, we develop new unsuper- vised learning models that can jointly detect emotions and summarize their triggers. Our best approach, entitled Emotion-Aware Pager- ank, incorporates emotion information from external sources combined with a language un- derstanding module, and outperforms strong baselines. We release our data and code at https://github.com/tsosea2/CovidET-EXT. 1 Introduction Language plays a central role in social, clinical, and cognitive psychology (Pennebaker et al., 2003), and social media presents a gold mine for such analysis: people turn to social media to share ex- periences around challenges in their personal lives and seek diagnosis, treatment, and emotional sup- port for their conditions (Choudhury and De, 2014; Gjurkovi´c and Šnajder, 2018). During crises, such as natural disasters or global pandemics, large-scale analysis of language on social media — both how people feel and what’s going on in their lives to lead to these feelings — can have a profound im- pact on improving mental health solutions as well *Tiberiu Sosea and Hongli Zhan contributed equally. Emotion Triggers in a Reddit Post 1. It finally happened. 2. Took an older relative for her first Pfizer dose. 3. Not that many people showed up so all accompanying family members were offered the shot and give papers for a second dose. 4. I wasnt due to get my shot for the next couple of months and have had some scares, many for which Ive gotten support from you lovely awesome people. 5. I now have a whole different perspective on my governments organization. 6. They have a good oiled vaccination machine. 7. They just need more doses. 8. Best part is I was told I could come back for my second dose whenever my relative was scheduled to get Hers. 9. I have a lil arm pain. 10. But its the slightest arm pain when moving it past certain angles. 11. Ive noticed that a lil blood drop shows on my vaccination spot (took a shower half an hour later). …

Content truncated to 500 words