Strategies for Collecting, Processing, Analyzing, and Preserving Tweets from Large Newsworthy Events


#WomensMarch, #Aleppo, #paris, #bataclan, #parisattacks, #porteouverte, #jesuischarlie, #jesuisahmed, #jesuisjuif, #charliehebdo, #panamanpapers, and #exln42 are all different hashtags, but they share several things in common. They are all large newsworthy events. They are datasets that each contain over a million tweets. Most importantly these collections raise some interesting insights in collecting, processing, analyzing, preserving large newsworthy events. Collecting tweets from these events can be challenging because of timing. Tweets can be collected from the Filter API and Search API. Both having their own caveats. The Filter API only captures the current Twitter stream, and is limited to collecting up to 1% of the overall Twitter stream. The Search API allows you to collect more than 1% of the overall Twitter stream, but one can only collect up to 18,000 every 15 minutes, and is limited to a 7 day window. Generally, using a strategy of using the Filter and Search API to capture a given event is the best. DocNow's twarc includes a number of utilities to process a dataset after collection. These tools allow a researcher, librarian, or archivist to filter their dataset(s) down to what is needed for appraisal, and then accession. Noteworthy tools include; deduplication, source, retweets, date/times, users, and hashtags. DocNow’s utilities can be further used to curate related collections. One can extract all the urls of a dataset, unshorten them, and extract the unique urls to use as a seed list for a web crawler to capture websites related to a given event. One can also extract all of the image urls, and download all images associated with a dataset, which then can be used for image analysis, presentation, and/or preservation.

Toronto, Canada