Extracting a Knowledge Base of COVID-19 Events from Social Media

Proceedings of the 29th International Conference on Computational Linguistics, pages 3810–3823 October 12–17, 2022. 3810 Extracting a Knowledge Base of COVID-19 Events from Social Media Shi Zong1 Ashutosh Baheti2 Wei Xu2 Alan Ritter2 1University of Waterloo 2Georgia Institute of Technology s4zong@uwaterloo.ca, abaheti95@gatech.edu {wei.xu, alan.ritter}@cc.gatech.edu Abstract We present a manually annotated corpus of 10,000 tweets containing public reports of five COVID-19 events, including positive and neg- ative tests, deaths, denied access to testing, claimed cures and preventions. We designed slot-filling questions for each event type and an- notated a total of 28 fine-grained slots, such as the location of events, recent travel, and close contacts. We show that our corpus can support fine-tuning BERT-based classifiers to automat- ically extract publicly reported events, which can be further collected for building a knowl- edge base. Our knowledge base is constructed over Twitter data covering two years and cur- rently covers over 4.2M events. It can answer complex queries with high precision, such as “Which organizations have employees that tested positive in Philadelphia?” We believe our pro- posed methodology could be quickly applied to develop knowledge bases for new domains in response to an emerging crisis, including natural disasters or future disease outbreaks.1 1 Introduction Since December 2019, the novel coronavirus rapidly spread across the world, and consequently, a flood of COVID-19 related information has ap- peared on social media. This includes reports on public figures who have tested positive/negative for the virus, which often break first on Twitter, such as Bill Gates’s announcement as shown in Figure 1. Besides public figures, individual users and orga- nizations on Twitter also report COVID-19 events around the world. For example in January 2021, many sources in different countries reported an in- creasing number of new cases exported from the UK (Figure 2). Being able to gather this informa- tion can potentially help experts and the general 1Our corpus (with user-information removed), automatic extraction models, and the corresponding knowledge base are publicly available at https://github.com/viczong /extract_COVID19_events_from_Twitter. Figure 1: Example tweet that contains a self-reported TESTED POSITIVE event. public to quickly identify issues and assess the situation near real-time, complementing officially reported data which may take longer to obtain, and does not include information at the same level of granularity as that reported in natural language on news and social media. In this paper, we present an empirical study on the extraction of large quantities of structured knowledge related to an ongoing pandemic from Twitter. To achieve this, we construct a corpus of 10,000 tweets with rich linguistic annotations, cov- ering five event types: positive tests, negative tests, denied access to testing, deaths, claimed methods of cure and prevention. More specifically, we an- notate fine-grained semantic information for each event type by designing slot-filling questions and asking annotators to highlight text spans as an- swers. We show that our corpus can support train- ing BERT-based classifiers to extract structured information automatically from Twitter. While slot F1 scores vary from 0.3 to 0.9 in individual tweets (most F1 scores …

Content truncated to 500 words