An Empirical Methodology for Detecting and Prioritizing Needs during Crisis Events

Findings of the Association for Computational Linguistics: EMNLP 2020, pages 4102–4107 November 16 - 20, 2020. c⃝2020 Association for Computational Linguistics 4102 An Empirical Methodology for Detecting and Prioritizing Needs during Crisis Events M. Janina Sarol, Ly Dinh, Rezvaneh Rezapour, Chieh-Li Chin, Pingjing Yang, Jana Diesner University of Illinois at Urbana-Champaign, IL, USA {mjsarol,dinh4,rezapou2,cchin6,py2,jdiesner}@illinois.edu Abstract In times of crisis, identifying essential needs is crucial to providing appropriate resources and services to affected entities. Social media plat- forms such as Twitter contain a vast amount of information about the general public’s needs. However, the sparsity of information and the amount of noisy content present a challenge for practitioners to effectively identify relevant information on these platforms. This study proposes two novel methods for two needs de- tection tasks: 1) extracting a list of needed resources, such as masks and ventilators, and 2) detecting sentences that specify who-needs- what resources (e.g., we need testing). We evaluate our methods on a set of tweets about the COVID-19 crisis. For extracting a list of needs, we compare our results against two of- ficial lists of resources, achieving 0.64 preci- sion. For detecting who-needs-what sentences, we compared our results against a set of 1,000 annotated tweets and achieved a 0.68 F1-score. 1 Introduction During crises, substantial amounts of information are shared and discussed on social media (Palen and Anderson, 2016; Reuter et al., 2018). Some of these posts may contain relevant information about the needs of affected and at-risk populations (Basu et al., 2018; Dutt et al., 2019; Purohit et al., 2014). The recent COVID-19 virus outbreak is no exception; online platforms such as Twitter have been crucial means for sharing information about the impact of the outbreak (Singh et al., 2020), per- sonal accounts from infected individuals (Jimenez- Sotomayor et al., 2020), and updates from medi- cal professionals (Rosenberg et al., 2020). Crisis responders and practitioners have also turned to online platforms to obtain actionable information that could aid them in response planning (Vieweg et al., 2010; Zade et al., 2018). In particular, schol- ars in crisis informatics have provided solutions to detect relevant Twitter messages that express resource needs and availabilities related to crisis events, e.g., during the 2015 Nepal earthquake (Basu et al., 2017; Dutt et al., 2019) and the 2015 Chennai floods (Sarkar et al., 2019). This paper builds upon and extends prior literature by propos- ing two needs detection tasks and applying needs detection to data about the COVID-19 crisis. In par- ticular, we (1) extract a list of needs by using word embeddings to identify closest terms to needs and supplies with respect to their cosine similarity, and (2) detect who-needs-what sentences to determine social entities who need particular resources. This study makes two contributions. First, we propose a method for identifying and prioritizing resource needs during a crisis. Second, we present a set of heuristics to determine the social entities that need specific resources. Overall, our study provides a reliable set of methods that might help response professionals …

Content truncated to 500 words