This is not a penis: User-generated tags

This is not a penis:Professional descriptors VS game-generated tags

Crowdsourcing for the Digital Humanities and Cultural Heritage Sector, 23 April 2013Rasmus Thøgersen

Human Computation

Games!!

Games With A Purpose (GWAP) “Rather than using computer vision techniques, which don’t work well enough, we encourage people to do the work by taking advantage of their desire to be entertained”

von Ahn & Dabbish, 2004

As an answer to

Image of Erik Kessel’s Photography In Abundance (2011) removed, Image of Erik Kessel’s Photography In Abundance (2011) removed,

Make a Difference

Comparison between

Descriptors• Created by professionals with

specialized knowledge of both the specific collection and the wider context.

• Based on some formal taxonomy

• Also fairly cost-intensive to ask a trained librarian to index ~2000 images (or 18 million)

• Only one perspective

Tags• Created by everyone• A folksonomy• Verified by other players• Cheaper• Probably flawed• Chaotic and free

RQ1

RQ2

RQ1:• More than 50% of the 3Vtags had a semantic relation to the Descriptors• Approx. a third of the 3Vtags had a same/equivalence relation• The second most common relation was the associative

RQ2:• The vast majority of 2Vtags and 3Vtags were Objects/Artifacts• The most interesting descriptions were found in the Free tags• Vandalism do exist…

?