[Paper Review] Is Working From Home The New Norm? An Observational Study Based on a Large Geo-tagged COVID-19 Twitter Dataset

Yunhe Feng, Wenjun Zhou|arXiv (Cornell University)|Jun 15, 2020

Sentiment Analysis and Opinion Mining5 references14 citations

TL;DR

This study uses a large geo-tagged Twitter dataset of 650,563 U.S. tweets (Jan 25–May 10, 2020) to analyze public sentiment and work engagement during the COVID-19 pandemic. It proposes hourly tweeting volume comparisons between workdays and weekends to estimate work-from-home trends, finding that New York showed lower work engagement initially, while afternoon work engagement rose sharply during reopening. The dataset and findings are publicly shared for further research.

ABSTRACT

As the COVID-19 pandemic swept over the world, people discussed facts, expressed opinions, and shared sentiments on social media. Since the reaction to COVID-19 in different locations may be tied to local cases, government regulations, healthcare resources and socioeconomic factors, we curated a large geo-tagged Twitter dataset and performed exploratory analysis by location. Specifically, we collected 650,563 unique geo-tagged tweets across the United States (50 states and Washington, D.C.) covering the date range from January 25 to May 10, 2020. Tweet locations enabled us to conduct region-specific studies such as tweeting volumes and sentiment, sometimes in response to local regulations and reported COVID-19 cases. During this period, many people started working from home. The gap between workdays and weekends in hourly tweet volumes inspired us to propose algorithms to estimate work engagement during the COVID-19 crisis. This paper also summarizes themes and topics of tweets in our dataset using both social media exclusive tools (i.e., #hashtags, @mentions) and the latent Dirichlet allocation model. We welcome requests for data sharing and conversations for more insights. Dataset link: http://covid19research.site/geo-tagged_twitter_datasets/

Motivation & Objective

• To understand how public sentiment and work behavior varied across U.S. states during the early COVID-19 pandemic.
• To develop a method for estimating work engagement using hourly tweeting volume differences between workdays and weekends.
• To examine the impact of local policies—such as stay-at-home orders and reopenings—on public sentiment and work patterns.
• To provide a publicly accessible, large-scale geo-tagged Twitter dataset for future pandemic-related research.

Proposed method

• Collected over 170 million English-language COVID-19-related tweets via Twitter’s Streaming API from January 25 to May 10, 2020.
• Filtered and cleaned data by removing tweets from detected bots, defined as users posting over 5,000 tweets or over 1,000 tweets with highly concentrated posting intervals.
• Extracted geo-tagged tweets by identifying tweets with 'country_code' = 'US' and matching state names from 50 states and Washington, D.C.
• Proposed a work engagement metric based on the difference in hourly tweeting volumes between workdays and weekends, focusing on 8:00–17:00.
• Applied hashtag and @mention analysis alongside Latent Dirichlet Allocation (LDA) to identify dominant topics in the dataset.
• Conducted sentiment analysis using polarized words and facial emojis to track emotional shifts around key pandemic events.

Experimental results

Research questions

• RQ1 How did public sentiment toward stay-at-home orders and reopening vary across U.S. states during the early pandemic?
• RQ2 To what extent did local COVID-19 case and death rates correlate with regional tweet volumes and sentiment?
• RQ3 How did work engagement, measured via hourly tweeting patterns, change in response to stay-at-home orders and reopening policies?
• RQ4 What were the dominant themes and topics discussed on Twitter in different U.S. regions during the pandemic?
• RQ5 How did public emotions evolve around key events such as the 100th confirmed case or 1,000th death?

Key findings

• Residents in Oregon, Montana, Texas, and California exhibited more intense reactions to confirmed cases and deaths compared to other states, as measured by normalized tweet volumes.
• New York state showed significantly lower work engagement in the first week of stay-at-home orders compared to other states, based on reduced midday tweeting activity.
• The average hourly work engagement in the afternoon (13:00–16:59) increased substantially during the first week of reopening compared to the first week of lockdown.
• Negative sentiment dominated public reactions to major pandemic milestones, such as the 100th and 1,000th confirmed cases and deaths, across all states.
• The dataset contains 650,563 unique geo-tagged tweets from 246,032 users, with 5.96% sourced from an external dataset, and is publicly available at http://covid19research.site/geo-tagged_twitter_datasets/.
• Only 0.055% of users posted more than one geo-tagged tweet per day on average, indicating broad geographic representation and minimal user bias.