#S17 Timeline: A Twitter Story
Using over 30,000 unique tweets collected during Sept 15th to Sept 18th extracted with the Python Pattern package and that contained keywords relating to NYC based protests or actions we looked for tweets of interest based on our focus area decided on day one, those with multimedia content that could be used to retell event histories, and by filtering with regular expressions and mongoDB (which we learned to use earlier at the 10AM skillshare with Tom!).
OWS Anniversary Timeline:
Since we did not see many cases where specific actors were named, but there were many tweets that related to state and space and had media attached that appeared in the collection as we were annotating it, at the beginning of day two we decided to create a Verite timeline of Occupy’s anniversary as told through Twitter accounts. Twitter timestamps allowed us to piece together a cohesive story. These methods in no way present a comprehensive, unbiased (our tweets were extracted with pre-determined selection criteria) or temporally correct (timestamps provide an approximate anchor, but it is hard to discern the exact time of an event) event history. However, we do feel it presents a remarkable way to move towards a participant documented history of events, as they are happening, and a new and exciting mode of social-mediated storytelling. Mining real-time tweets helps to keeping the tale as close to the `true’ event, or its perception of the real event by occupiers as possible. It reduces the potential for erroneous (romanticized, sensationalized, vilified, or incomplete) translations that are more likely to occur, or harder to fact-check as the event becomes further displaced in time.
Day one we decided to focus on identifying incidents of police misconduct that identified the actors, but the task was more challenging than anticipated. However, we noted several other types of events related to the #S17 protests in NYC documented in Tweets and that we could visualize with word frequency techniques and used Many Eyes. We found four themes related to state and space we could aggregate and visualize:
Figure 1. Word cloud visualizing Winski, badge numbers, and the arrest of NY Councilman Junmaane Williams.
- involving Winski, a high-ranking officer that was associated with numerous tweets citing misconduct or intimidation.
- documenting the arrest and assault of Councilman Williams using a police baton, and
- mentioning badge numbers or the ranks of officers accused of misconduct, some of which we were able to corroborate with additional media evidence.
- noting journalist arrests, and reported with the hashtag #journarrest,
Figure 2. #journarrest tweets on S17
Figure 1. shows a word cloud to visualization of the first three types of tweets we collected. Figure 2. shows a word tree of Tweets documenting incidents with “white shirts”. Lastly, Figure 3. shows a word tree that branches from the term “white shirt” and visualizes the Tweet context surrounding the term.
Despite over 100 arrests on #S17, we did not collect as many incidents with named actors as expected. We attribute the low number of incidents relative to our last assessment to several possible reasons:
- the data had a shorter time span that our previous analysis and focuses primarily on one day, #S17
- a decrease in police misconduct
- officers are concealing their ids, or
- activists are not recording tweets in an obvious way.
One major limitation of this work is the quality of out data. For example, we encountered different depictions of the assault on Councilman Williams by the NYPD and the type of officer involved. Also, the credentials provided by journalist covering protests and actions can differ. Individuals labeled as journalists in tweets, may be a streamer without press credentials, and not an official press badge. It is unclear type of status the individual being arrested may possess, and if they possess press credentials. When possible, we tried to find additional evidence to validate Tweets.
Figure 3. White Shirts Officer Incidents
Out ongoing work will involve continuing to document key events using this data, and enhancing the timeline. Also, we will add the five tweets we did identify to out incident database.