Large-Scale COVID-19 Twitter Chatter for Scientific Research

Over 915 million unique tweets, including retweets, were collected from the Twitter Stream, with dedicated gathering starting March 11th yielding over 4 million tweets daily. The dataset includes a cleaned version without retweets, covers multiple languages with higher prevalence of English, Spanish, and French, and provides pre-processed n-gram frequencies for NLP tasks. It offers longitudinal coverage from at least January 27th, with specific additions like Russian language tweets collected between January 1st and May 8th.

Use Cases

• Tracking tweet dissemination and influence based on the inclusion of retweets.
• Multilingual sentiment or topic analysis based on the text of tweets and provided language metadata.
• Studying temporal patterns of public discussion using daily statistics and longitudinal coverage.
• Analyzing hashtag, mention, and emoji trends from the provided daily frequency files.
• Training or benchmarking NLP models for pandemic-related text using the provided frequent terms, bigrams, and trigrams.

Strengths

• Large scale, with one source reporting over 915 million unique tweets in the full dataset.
• Includes longitudinal coverage from at least January 27th, with daily collection volumes specified.
• Provides cleaned data (no retweets) and auxiliary files like n-gram frequencies to support analysis.
• Multilingual coverage with language and place location metadata included for all tweets since version 20.

Limitations

• Sources conflict on the total row count (776M vs. 915M unique tweets), indicating version discrepancies.
• Specific column schema is not detailed in the provided metadata, limiting precise understanding of data structure.
• The most recent update date is unknown, making it difficult to assess dataset freshness.

Provenance

Source

Juan M. Banda (Georgia State University) and collaborators including Katya Artemova (NRU HSE) and Elena Tutubalina (KFU).

Collection Method

Acquired from the Twitter Stream, with dedicated gathering starting March 11th.

Time Range

From at least January 27th, with specific Russian language data from January 1st to May 8th. Collection ongoing with daily yields reported.

Geography

Global, as data captures all languages from the Twitter Stream. Place location metadata is included.