A conceptual framework for studying reactions to events in location-based social media

Overview of paper progress and conducted analyses

1) Timeline

2) Analyses

a) St. Jude

Starting point: Storm-related Flickr images 2007 - 2017 filtered for the following labels in tags or descriptions: storm OR cyclone OR gale OR gust OR hurricane OR blow OR wind OR windy OR orkan ..compared to the overall number of available Flickr images per day for the timespan October-20 to November-10. (21 days)

1) Area: Europe; Timespan: xxxx-10-20 to xxxx-11-10 (21 days)
stjude_temporal.png

  • On St. Jude day of peak impact (Oct-28), almost 5% of all active Flickr users took storm-related pictures

2) Area: UKCoast (10km buffer), UK, France; Timespan: 2013-10-20 to 2013-11-10 (21 days)
stjude_spatial.png

  • Spatial filtering increases percentages: most storm-related pictures taken on coasts (+ earlier reactions)
  • possible bias for non-english speaking countries (France)

Data Statistics Flickr (Europe):

  • storm OR cyclone OR gale (xxxx-10-20 to xxxx-11-10 - 21 days):
  2011 2012 2013 2014 2015
photos 272 339 1036 237 96
users 108 184 362 145 68
tags 3917 4574 14042 3704 1653
unique tags 1094 1766 4381 1437 837
  • gust OR hurricane OR blow (xxxx-10-20 to xxxx-11-10 - 21 days):
  2011 2012 2013: 2014: 2015:
photos 63 41 49 65 12
users 10 19 27 20 11
tags 1091 506 792 768 141
unique tags 133 281 426 220 134
  • wind OR windy OR orkan (xxxx-10-20 to xxxx-11-10 - 21 days):
  2011 2012 2013: 2014: 2015:
photos 441 454 526 319 240
users 104 143 221 143 88
tags 6817 8381 10778 5771 4214
unique tags 1331 1954 3216 1517 1352

b) Brexit

Eva-Brexit-Analysis

TwitterBrexitReaktionen_Eva.pdf

Alex-Brexit-Analysis:

1a) Tweets from and @Nigel_Farage; Timespan: Thu 23 18:00 to Fri 24 9:50 2324_atFarage.png

  • Overall Sentiment is calculated based on all tweets with the Hashtag #EUref or
    #Brexit containing a clear negative or positive ‘sentiment’, based on Twitter’s
    advanced search (i.e. positive tweets will turn up posts including a range of common
    smilies, indicating the overall tone of the tweet was positive. Looking for negative
    tweets conversely fi nds tweets featuring a range of frowns).
    From 06/23/2016 18:00 to 06/24/2016 9:50, for hashtags #EUref or #Brexit, there
    were 4042 tweets with positive sentiment (smile) and 2580 tweets with negative
    sentiment (nosmile).
  • Smile (blue) and Nosmile (red) was calculated based on all tweets that
    are directed @Nigel_Farage with a clear expression of sentiment.
    From 06/23/2016 18:00 to 06/24/2016 9:50, this data encompasses 275
    positive tweets and 38 negative tweets (ratio = 7.23).
    Tweets from Nigel_Farage are attached in chronological order to the
    chart above

1b) Tweets from and @David_Cameron; Timespan: Thu 23 18:00 to Fri 24 9:50 2324_atCameron.png

  • Smile (blue) and Nosmile (red) was calculated based on all tweets that
    are directed @David_Cameron with a clear expression of sentiment.
    From 06/23/2016 18:00 to 06/24/2016 9:50, this data encompasses 147
    positive tweets and 94 negative tweets (ratio = 1.56).
    Tweets from David_Cameron are attached in chronological order to the
    chart above.

2a) Tweets from and @Nigel_Farage; Timespan: Thu 21 2016 to Fri 25 2016 2125_atFarage.png

  • From 06/21/2016 0:00 to 06/25/2016 0:00, for hashtags #EUref or #Brexit, there
    were 8512 tweets (of those, 191 Geotweets/ 2.2%) with positive sentiment (smile)
    and 4052 tweets (of those, 126 Geotweets/ 3.1%) with negative sentiment (nosmile).
  • Smile (blue) and Nosmile (red) was calculated based on all tweets that
    are directed @Nigel_Farage with a clear expression of sentiment.
    From 06/21/2016 0:00 to 06/25/2016 0:00, this data encompasses 432
    positive tweets and 57 negative tweets (ratio = 7.58).
    Tweets from Nigel_Farage are attached in chronological order to the
    chart above.

2b) Tweets from and @David_Cameron; Timespan: Thu 21 2016 to Fri 25 2016 2125_atCameron.png

  • Smile (blue) and Nosmile (red) was calculated based on all tweets that are directed @David_Cameron with a clear expression of sentiment. From 06/21/2016 0:00 to 06/25/2016 0:00, this data encompasses 313 positive tweets and 137 negative tweets (ratio = 2.29). Tweets from David_Cameron are attached in chronological order to the chart above.

[ ] Add comparison to overall #Brexit & #EURef tweets (Context)

c) Cherry Blossoming

1) Comparison of Reactions across Flickr / Twitter (AD)

Starting point: Cherry tweets or Flickr tags/titles Query: (Cherry AND (*blossom* OR *flower*)) OR (Sakura AND (*blossom* OR *flower*))

1a) Temporal Facet Twitter: cherry_0717_twitter.png

  • Overall rise & decline of Twitter network (=context/bias)
  • regular, reoccuring pattern; underlying contstant volume of noise

1b) Temporal Facet Twitter & Flickr: cherry_0717_twitter_flickr.png

  • Overall rise & decline of Twitter (later peak) & Flickr (earlier peak) network (=context/bias)
  • regular, reoccuring pattern in both networks; less noise for Flickr geo-data

1c) Spatial Facet Flickr: Worldwide compared to Washington DC Tidal Basin cherry_0717_flickr_WW_DCTB.png

  • Filtering for spatial facet will lead to a more clearly delimted event start/end times

2) Analysis of Flickr Cherry Blossoming Imagery in USA (G&NA)

map_all.png

  • map_all.png shows a spatial footprint of all photos 2007-2015 (we’ve
    excluded photos with dates before 2007; some of them are from 1947)

h2d_all.png

  • h2d_all.png: weekly counts of distinct flickr user IDs for weeks and
    years
  • Next, we’ve applied ST clustering using the method described in
    http://dx.doi.org/10.1109/DSAA.2015.7344880

ST clustering:

  • map_ST_clusters.png

  • ST_cube_all.png

  • STC_ST_clusters.png

Timeline ST clusters:

  • 2007 - 2009
    timeline_ST_clusters_2007_2009.png

  • 2010 - 2012
    timeline_ST_clusters_2010_2012.png

  • 2013 - 2015
    timeline_ST_clusters_2013_2015.png

Spatio-temporal clustering (our algorithm) parameters:
mini-cluster radius 100 km
maximal time gap 3 days
merge mini-clusters with at least 5 common events
ignore clusters with <10 events or <5 distinct event rources (UserID)

  • Result: 147 ST clusters; sizes from 10 to 7968 (DC)
    10 largest clusters with sizes from 1054 to 7968 are in DC
    N distinct users: from 5 to 251
    10 clusters with more than 100 distinct users are all in DC

Histograms:

  • All:
    h2d_all.png
  • Boston:
    h2d_Boston.png
  • DC and Philadelphia:
    h2d_DCandPhiladelphia.png
  • Los Angeles:
    h2d_Los_Angeles.png
  • New Jersey:
    h2d_New_Jersey.png
  • Oregon:
    h2d_Oregon.png
  • San Francisco:
    h2d_San_Francisco.png
  • Seattle:
    h2d_Seattle.png
  • Toronto:
    h2d_Toronto.png
  • Vancouver:
    h2d_Vancouver.png
  • NE:
    h2d_NE.png
  • NW:
    h2d_NW.png
  • h2d_ST_clusters:
    h2d_ST_clusters.png
  • After eliminating noise (occasional photos without sufficient number
    of neighbors in ST windows), we’ve built 2d histograms (weeks x years) of
    weekly counts of distinct flickr user IDs in different parts of the country.