@LitPlus Twitter Urban Sensorium [#locative #gps #geotag #urbanphenomenology]

Team Members: Pehr Hovey, Allison Schifani, Dana Solomon

 

 

 

Abstract:

 

 

We will cull location-specific Twitter updates or "tweets" (140-character limit information posted by users that are then distributed throughout user specific social networks online or via SMS) using sensory descriptors in verb form [see, hear, touch, taste, smell].  We will be limiting our search results to those generated within a certain radius of either Los Angeles or New York City.  

 

We will then try to examine any emergent patterns of comparative interest between the tweets produced in each location. 

 

By developing a method to aggregate small, user-authored texts (in the form of updates) and link them to specific locations (using gps information), we hope to learn something about the kind of communication enabled by Twitter as a text-specific social networking site, in real space and time. Using the data we collect, we hope that we can produce a comparative visualization that may give us insight into how individuals on a micro level and social populations on a macro level produce collective and interactive narratives. Because Twitter is so distinctly multivocal, it provides an ideal platform to engage questions of space, place, time and text in the contemporary moment.  

 

 

Background Theory and Research Questions:

 

"Linking both sides of these corridors, which get their light from above, are the most elegant shops, so that the passage is a city, a world in miniature" (Benjamin, 3).

 

"If this new multinational downtown effectively abolished the older ruined city fabric which is violently replaced, cannot something similar be said about the way in which this strange new surface in its own peremptory way renders our older systems of perception of the city somehow archaic and aimless, without offering another in their place?" (Jameson, 14)

  

Theory: One of the primary texts we have been thinking about is Benjamin's Arcades Project, a massive collection of exposes and recorded wanderings that helped to form and influence our understanding of the modern flaneur.  The fragmented nature of the work mirrors the flaneur's own fleeting interactions with urban ephemera, his or her "anamnestic intoxication [that] ... feeds on the sensory data taking shape before [the] eyes" (Benjamin, 417). The emphasis on the non-linear, constellated structure of the Arcades Project fit well with our object of analysis, the short, sometimes even truncated Twitter update.  Other relevant theorists include Henri Lefebrvre, Rem Koolhas, Anthony Vidler, David Harvey, and Fredric Jameson, particularly his assertion regarding the transformation of human perception brought on by postmodern spaces.  We are quite interested in the way in which mediated perception of urban space might impact user language and possibly even contribute to a shift in observation practices (i.e. what individuals notice or miss in an urban setting).  For more on Benjamin and the Arcade's Project, please refer to this Annotated Bibliography. 

 

Research questions: How does a conversation or narrative, engaged en masse, unfold in space and time? How might locative media, such as gps-enabled twitter updates alter the way individuals perceive urban space, vis a vis the 5 senses? What can alternative visualizations of this kind of narrative construction tell us about conversation flows and counterflows? What blocks do we come up against in engaging this information, both technical and conceptual? What could potential (comparative) visualizations tell us about how this form of micro-writing functions differently in different spaces, or in relation to space more broadly? Finally, might visualizing data in interesting ways contribute to a transformation in the way we, as researchers, perceive it? 

 

 

Process/Code:

 

1.) Select Individual Sensory Descriptor in Simile Structure (smells like, sounds like, tastes like, etc.)

2.) Filter Data Set for GPS-enabled updates (including radius from desired starting location)

3.) Map Hash Tag using traditional mapping software and existing programs (Google Maps/Google Earth)

4.) Produce Alternative Visualization of final data set

5.) Repeat 1-5 for each selected location

6.) Interpret comparative visualizations

 

Twitter search examples:

 

Find tweets near Santa Barbara

http://search.twitter.com/search.atom?geocode=34.412334%2C-119.845932%2C50mi

• data gathering / processing 
  • searches run every 5 minutes
  • data processing step
    • geocoding
    • emoticons, hashtags, mentions, urls
    • emotion (currently only emoticon)
    • readability (fog, flesch/kincaid)
    • time-of-day
    • 'readable text'
• limitations of geotagging
  • less than 1% are geotagged
  • 90%+ have user location that can be geocoded
  • dealing with ambiguous data
• future data processing
  • better emotions (text analysis)
  • built-in concordance / POS tagger ...analyze  A looks like  B

 

 

Visualizations: 

 

 

 

 

 

 

 

Results and Findings:

 

     We compared New York City tweets to Los Angeles tweets using a handful of freely available online analysis and visualization tools. Though most of the TaPor tools remained offline during the course of our project, the word list tool which counts frequency of words on at a given URL was still functional and so, initially, we pasted the text of 10,000 New York tweets on a blog (as the only post) and had TaPor process the url. We then replaced the post with 10,000 Los Angeles tweets and repeated the process. In these samples, the two cities did not produce a particularly marked difference in number of occurences of a given sensory simile, though there were indeed differences. What was perhaps more interesting about the data was the preference users had for visual similes over the other four sensory similes we searched for in both cities. "Looks like" occured nearly three times more frequently than "feels like" in both LA and NY tweets, and nearly double the frequency of "sounds like."  "Smells like" and "tastes like" occured quite rarely in the sets, but the most marked difference between the two cities could be found in those similes. Los Angeles users included "smells like" in 223 tweets where NY users used it only 14 times. LA tweets also included "tastes like" 13 times more than did New Yorker tweets. We uploaded samples of our data onto IBM's ManyEyes visualization creation platform. By far the most successful visualizations, at least aesthetically, were the Phrase Nets and Word Trees. They are also the most dynamic, allowing you to explore the data, change perameters, search for phrases, etc. There are a number of problems, however. In these visualizations, the program treats all the tweets together, as a single text. It cannot recognize separate tweets so when you explore the visualizations, the trees in particular, you may have false links between words. The other problem is the sets of data we were able to upload didn't match particularly well with the requirements of ManyEyes: when values are not numeric (for example our search terms, the sensory similes) the program often cannot process the data at all. Going through ManyEyes also means the visualizations do not change in real time. You upload a single set of data, and cannot pipe in new data as it comes. It could not process emoticons either, or symbols--both very common among Twitter users. What these few visualizations through ManyEyes do do, certainly, though is confirm something of what we had anticipated, not about the data, but about the project of visualizing it: looking at the content of the data in alternative ways does change the way you are able to grasp it. In the tag clouds and word trees we can see the weight of a word in the set. It may tell me something that LA twitter users included the abbreviation "lol" in 938 of the 10,000 tweets we analyzed, but to see the various pathways it leads to, to see it next to other words with which it was paired, to be able to follow it in a number of trajectories, or lines of flight? That is an altogether different sense of how it links to experience. There may be something about this specific to the terms of our search. We were attempting to trace ways in which urban micro-blogging is engaged with everyday experience. Data visualization are specifically about experience--visual and otherwise. Numbers in a raw data set are not.

 

 

Memorable Tweets: