[Asis-l] tracing topics over a half decade of american television news

[kalev leetaru]

I thought many of you would find of great use my latest project, debuting
today, which provides a range of analyses and visualizations against the
Internet Archive's Television News Archive:


http://blog.gdeltproject.org/television-trends-tracker-explore-american-television-news/
http://analysis.gdeltproject.org/cgi-bin/iatvtimeline/iatvtimeline


You can specify up to three words or phrases and optionally limit to a
particular television network or date range, and get back a fully
interactive dashboard showing how your terms have been used on American
television news over the past half-decade (as captured by the IA Archive).
 See how coverage volume has changed by week, which television networks
devote the most attention, the topics that feature more prominently in
shows mentioning the terms, and the words that appear most frequently in
the immediate context of the terms.  It even features a contextual ngrammer
that computes a word histogram of all of the words that appear in the
immediate vicinity of your term(s) in the closed captioning stream, making
it possible to see how your term(s) are being contextualized in the news.

Of particular interest for political advertising, under the network
dropdown there is a special option to select all networks in the
Philadelphia area as part of the Archive's Philadelphia Area Political Ads
Pilot Project (https://archive.org/details/PHL).

You can either view the results online in your browser or download to an
Excel spreadsheet for further analysis with various content analysis or
statistical software.

This is a prototype service that will be refined over time, designed to
provide a first attempt at an in-depth research exploration interface to
the Archive's holdings, so would love any feedback or other thoughts.  All
kinds of fascinating possibilities re being able to measure how various
issues are being discussed on television, especially political discourse
with the upcoming election season and the Philadelphia archive (which
itself is an incredibly rich resource).  The results are currently
unnormalized (they reflect the raw coverage volume recorded by the Archive)
and we are exploring a number of normalization approaches.

The contextual ngrams are particularly powerful for their ability to allow
in-depth analysis at how contextualization of a term or topic has changed
over time or across networks (or how similar/related terms differ in their
contextualization).  You can use the network dropdown and the Excel output
option to generate the full word histogram for your term for each network
and then compare.  Or, perhaps most powerfully, use the date selector to
limit by month over the last five years and output the Excel sheet for each
month and import those ngrams into statistical software or a content
analysis package for coding and further analysis of how the context around
a term is changing over time (being careful to accommodate for the fact
that some stations are added later than others and there are some sporadic
outages of some stations in the archive - you might narrow by network for
this).

Thought this would be of great interest and use to many of you!

Kalev Leetaru
2013-2014 Yahoo! Fellow, Georgetown University
http://kalevleetaru.com/
http://gdeltproject.org/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.asis.org/pipermail/asis-l/attachments/20140915/06eb17c9/attachment-0001.html>