EnsyNet: A Dataset for Encouragement and Sympathy Detection

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 5444–5449 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 5444 ENSYNET: A Dataset for Encouragement and Sympathy Detection Tiberiu Sosea Cornelia Caragea Computer Science University of Illinois Chicago tsosea2@uic.edu cornelia@uic.edu Abstract More and more people turn to Online Health Communities to seek social support during their illnesses. By interacting with peers with similar medical conditions, users feel emotionally and socially supported, which in turn leads to better adherence to therapy. Current studies in Online Health Communities focus only on the presence or absence of emotional support, while the available datasets are scarce or limited in terms of size. To enable development on emotional support detection, we introduce ENSYNET, a dataset of 6, 500 sentences annotated with two types of support: encouragement and sympathy. We train BERT-based classifiers on this dataset, and apply our best BERT model in two large scale experiments. The results of these experiments show that receiving encouragements or sympathy improves users’ emotional state, while the lack of emotional support negatively impacts patients’ emotional state. We make our code and data available at https://github.com/tsosea2/EnSyNet. Keywords: Sympathy, Encouragement, Online Health Communities 1. Introduction Online Health Communities (OHCs) have become a major source of support for people with health prob- lems and for their caregivers. People usually seek social support during their treatments (LaCoursiere, 2001), and such support is a key component to their overall well-being, having the potential to directly af- fect the success of their treatment (Coulson et al., 2007; Kim et al., 2004). Social support in OHCs can be di- vided in two core components: emotional and infor- mational support (Olson, 1995; Gooden and Winefield, 2007; Eysenbach et al., 2004; Brennan and Ripich, 1994). These two core functions of OHCs help to pro- mote new treatments, lifestyle, and can also reveal ad- verse drug effects (Yang et al., 2012). In this paper, we focus on emotional support detec- tion, more specifically on the expression of encour- agements and sympathy. Encouragement is the action of giving someone support, confidence, or hope, while sympathy is the feeling of pity or sorrow for someone else’s misfortune. These two types of support are ex- tremely beneficial for patients in improving their emo- tional state. To our knowledge, we are the first to ad- dress this fine-grained approach, as previous studies on emotional support detection focused only on a high level view of emotional support, which analyzes the presence or absence of such support (Khanpour et al., 2018; Biyani et al., 2014; Wang et al., 2014). To this end, we introduce ENSYNET, a new dataset composed of 6, 500 sentences annotated with encouragement and sympathy. ENSYNET is a challenging benchmark, be- ing more fine-grained and 6 times larger than other datasets used for detecting emotional support (Biyani et al., 2014; Wang et al., 2014). To indicate a few chal- lenges of the task, we show some examples from our dataset in Table 1. For …

Content truncated to 500 words