Knowing Funny: Genre Perception and Categorization in Social Video Sharing

Jude Yew, David Ayman Shamma, Elizabeth F. Churchill
CHI 2011

Categorization of online videos is often treated as a tag suggestion
task; tags can be generated by individuals or by machine
classification. In this paper, we suggest categorization
can be determined socially, based on people’s interactions
around media content without recourse to metadata that are
intrinsic to the media object itself. This work bridges the gap
between the human perception of genre and automatic categorization
of genre in classifying online videos. We present
findings from two internet surveys and from follow-up interviews
where we address how people determine genre classification
for videos and how social framing of video content
can alter the perception and categorization of that content.
From these findings, we train a Naive Bayes classifier to predict
genre categories. The trained classifier achieved 82%
accuracy using only social action data, without the use of
content or media-specific metadata. We conclude with implications
on how we categorize and organize media online
as well as what our findings mean for designing and building
future tools and interaction experiences.