Exploratory Analysis of Covid-19 Tweets using Topic Modeling, UMAP, and DiGraphs (2005.03082v1)

Abstract: This paper illustrates five different techniques to assess the distinctiveness of topics, key terms and features, speed of information dissemination, and network behaviors for Covid19 tweets. First, we use pattern matching and second, topic modeling through Latent Dirichlet Allocation (LDA) to generate twenty different topics that discuss case spread, healthcare workers, and personal protective equipment (PPE). One topic specific to U.S. cases would start to uptick immediately after live White House Coronavirus Task Force briefings, implying that many Twitter users are paying attention to government announcements. We contribute machine learning methods not previously reported in the Covid19 Twitter literature. This includes our third method, Uniform Manifold Approximation and Projection (UMAP), that identifies unique clustering-behavior of distinct topics to improve our understanding of important themes in the corpus and help assess the quality of generated topics. Fourth, we calculated retweeting times to understand how fast information about Covid19 propagates on Twitter. Our analysis indicates that the median retweeting time of Covid19 for a sample corpus in March 2020 was 2.87 hours, approximately 50 minutes faster than repostings from Chinese social media about H7N9 in March 2013. Lastly, we sought to understand retweet cascades, by visualizing the connections of users over time from fast to slow retweeting. As the time to retweet increases, the density of connections also increase where in our sample, we found distinct users dominating the attention of Covid19 retweeters. One of the simplest highlights of this analysis is that early-stage descriptive methods like regular expressions can successfully identify high-level themes which were consistently verified as important through every subsequent analysis.

Exploratory Analysis of Covid-19 Tweets Using Topic Modeling, UMAP, and DiGraphs

This paper, authored by Ordun, Purushotham, and Raff, presents an exploratory paper of COVID-19-related tweets utilizing a variety of analytical techniques, including pattern matching, Latent Dirichlet Allocation (LDA), Uniform Manifold Approximation and Projection (UMAP), and network visualization through directed graphs. This paper aims to elucidate the distinctiveness and prevalence of topics, assess the speed of information dissemination, and analyze network behaviors associated with COVID-19 discourse on Twitter.

Analysis Techniques and Findings

1. Keyword and Topic Analysis: The authors employ multiple methodologies to gain insights into the topics discussed in COVID-19 tweets. Using pattern matching and LDA, twenty topics were identified, with emerging themes noted around healthcare workers, personal protective equipment (PPE), and government announcements. Thematic upticks related to U.S. cases were observed post-White House briefings, highlighting the impact of governmental communication on public discourse.
2. Utilization of UMAP: UMAP was applied as a novel approach within the context of COVID-19 Twitter data. This dimensionality reduction and visualization tool facilitated the identification of unique clustering behaviors amongst topics, providing a method to assess the quality and distinction of the topics produced by LDA. The visualization effectively demonstrated distinct separations of themes, validating the interpretability and coherence of identified topics such as PPE and healthcare worker discussions.
3. Retweeting Time Analysis: The paper provides a quantitative analysis of retweeting speeds, finding a median retweeting time of 2.87 hours for COVID-19 tweets. This is notably faster than previously reported retweet times for other health-related information such as the H7N9 outbreak, indicating a heightened sense of urgency and a larger global audience during the COVID-19 pandemic.
4. Network Behavior and Retweet Cascades: Utilizing directed graph models, the paper visualizes retweet cascades over time, demonstrating how network density increases as retweeting time expands. The analysis identifies distinct users who dominate COVID-19 discourse, offering insights into the formation and influence of information networks on Twitter. The dynamic changes in these networks highlight how information propagation can lead to concentrated attention on specific users over longer periods.

Implications and Future Directions

The research provides a comprehensive view of Twitter dynamics during the COVID-19 pandemic, using advanced machine learning and network analysis techniques that offer insights into public response and the dissemination of information. These findings have significant implications for public health communication strategies, emphasizing the critical role of timely and clear messaging from government and health organizations.

In terms of future research, the methods applied could be extended to analyze longer and continuous streams of Twitter data, providing a more refined understanding of evolving topics and sentiment over time. Moreover, the visualization techniques, particularly UMAP and network graphs, hold potential for analyzing other types of online discourse, fostering more robust strategies in responding to public health crises and managing information flow.

Overall, this paper contributes a technical and methodologically diverse approach to understanding the complex dynamics of social media discourse around global health crises, paving the way for enhanced strategic communication and analysis frameworks in the field.