[Paper Review] Extracting COVID-19 Events from Twitter

Shi Zong, Ashutosh Baheti+2 more|arXiv (Cornell University)|Jun 3, 2020

Misinformation and Its ImpactsSocial Sciences9 references22 citations

TL;DR

This paper introduces a manually annotated corpus of 7,500 Twitter tweets labeled with COVID-19 events such as positive test results and denied testing access. Using this corpus, the authors demonstrate effective automatic extraction of event mentions through slot-filling, enabling structured representation of self-reported pandemic-related experiences, with tools and data made publicly available.

ABSTRACT

We present a corpus of 7,500 tweets annotated with COVID-19 events, including positive test results, denied access to testing, and more. We show that our corpus enables automatic identification of COVID-19 events mentioned in Twitter with text spans that fill a set of pre-defined slots for each event. We also present analyses on the self-reporting cases and user's demographic information. We will make our annotated corpus and extraction tools available for the research community to use upon publication at this https URL

Motivation & Objective

• To create a high-quality, manually annotated corpus of Twitter posts reporting COVID-19-related events.
• To enable automatic identification of specific event types—such as positive test results and denied testing access—using text spans in predefined slots.
• To analyze self-reported cases and user demographic patterns in social media during the pandemic.
• To support future NLP research by releasing the annotated corpus and extraction tools to the public.

Proposed method

• Manual annotation of 7,500 tweets with predefined event types and slot-filling structures for each event.
• Definition of a set of pre-defined event slots (e.g., event type, date, location) to standardize event representation.
• Application of sequence labeling and event extraction techniques to identify spans in text corresponding to each slot.
• Use of statistical and qualitative analysis to examine patterns in self-reported cases and user demographics.
• Development of tools for event extraction based on the annotated corpus for downstream use.

Experimental results

Research questions

• RQ1 How accurately can event types such as positive test results and denied access to testing be identified in social media text?
• RQ2 To what extent do self-reported cases in Twitter align with official case data or demographic trends?
• RQ3 What patterns emerge in user demographics among individuals reporting COVID-19 events on Twitter?
• RQ4 Can a structured slot-filling approach effectively extract meaningful event information from informal social media language?

Key findings

• The annotated corpus enables accurate automatic identification of COVID-19 events through slot-filling, supporting structured event extraction.
• A significant number of self-reported cases were found in the corpus, reflecting real-time public health reporting via social media.
• Demographic patterns in user reports suggest variations in reporting behavior across different user groups.
• The corpus and associated tools are publicly released to support ongoing research in NLP and public health informatics.