[论文解读] Exploratory Analysis of Covid-19 Tweets using Topic Modeling, UMAP, and DiGraphs

Catherine Ordun, Sanjay Purushotham+1 more|arXiv (Cornell University)|May 6, 2020

Misinformation and Its Impacts被引用 62

一句话总结

本研究应用 LDA 主题模型、UMAP 可视化和基于 DiGraph 的转发网络，对大型 Covid-19 Twitter 语料库进行分析，揭示不同的主题聚类、信息传播速度模式以及政府公告对推文活跃度的影响。

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.

研究动机与目标

• Investigate high-level trends in Covid-19 tweets and public discourse.
• Identify events that trigger spikes in Twitter activity about Covid-19.
• Determine how distinct topics are from each other within the corpus.
• Assess the speed of information dissemination via retweets compared to prior outbreaks.
• Analyze network dynamics of Covid-19 retweet cascades over time.

提出的方法

• Collect a large Covid-19 Twitter dataset using the Streaming API from March 24 to April 9, 2020 (23,830,322 tweets).
• Preprocess text by removing retweets, normalizing text, and extracting 13 keyword terms for trend analysis.
• Apply TF-IDF vectorization (max_features=10000) and LDA (20 topics, coherence 0.344) to generate topics for interpretation.
• Use UMAP to project TF-IDF features and topic labels into a 2D visualization to assess topic clustering and quality.
• Compute time-to-retweet metrics from retweet metadata to measure dissemination speed (median 2.87 hours; mean 12.3 hours).
• Construct and analyze retweet networks as nine time-point directed graphs to study cascade structure and density.

实验结果

研究问题

• RQ1 What high-level trends can be inferred from Covid-19 tweets?
• RQ2 Are there events that lead to spikes in Covid-19 Twitter activity?
• RQ3 Which topics are distinct from each other?
• RQ4 How does the speed of retweeting in Covid-19 compare to other emergencies and outbreaks?
• RQ5 How do Covid-19 networks behave as information spreads?

主要发现

• A 20-topic LDA model (coherence 0.344) identifies themes such as PPE, healthcare workers, and case/death mentions, with several non-English topics (Spanish, Portuguese, Italian, French).
• UMAP visualization of TF-IDF+LDA topics reveals distinct topic clusters and a separate “100: N/A” topic indicating unique tweet content.
• Live White House Coronavirus Task Force briefings trigger upticks in the topic corresponding to government announcements (Topic 18: potus).
• Median retweet time for Covid-19 messages in March 2020 is 2.87 hours (mean 12.3 hours), about 50 minutes faster than H7N9 reposting on Sina Weibo in March 2013.
• Retweet cascade analysis shows increasing network density as retweet speed rises, with nine time-point graphs illustrating fast-to-slow propagation and dominant users.
• Rapid retweeters often share content containing URLs (TF-IDF features include many URL-related terms), and user descriptions suggest political or news-focused accounts.