Community-Contributed Media Collections: Knowledge at Our Fingertips - Community-Built Databases: Research and Development - page 29

much less predictable evolution trends. Furthermore, the popularity of UGC pre-
sents rather diverse and complex dynamics, thus making traditional content popu-
larity predictions unsuitable.
Works devoted to studying media content distribution address either (a) the
lifespan of the media content or (b) social popularity as shown in Fig. 2.1 . The
lifespan of the media content is highly dependent on the user behavior and interests.
Several studies have been investigating the main reasons for the popularity and
diffusion of videos and photos.
For example, in [ 18 ] the analysis of the popularity evolution of user-produced
videos in YouTube and other similar UGC Web sites is presented. The key
observation is that understanding the popularity characteristics can prove pivotal
in discovering weaknesses and bottlenecks in the system and suggest policies to
improve it. The study was conducted on several datasets containing video meta
information crawled from YouTube and Daum [ 21 ], a popular search engine and
UGC service in Korea. Their analysis reveals that the popularity distribution of
videos exhibits power-law behavior with a steep truncated tail (exponential cut-
off), suggesting that requests for videos are highly skewed toward popular files.
Rather than this skewness being a natural phenomenon due to the low level of
interest in many UGC videos, filtering effects in search engines, which typically
favor a small number of popular items, seem most likely responsible for the
significant imbalance in the video popularity distribution. Popular videos thus
tend to gain more and more views, while niche videos reach a much smaller
audience than expected. Proper leverage of the latter could increase the total
number of views by as much as 45% and reveal the latent demand created by the
search engine bottleneck.
Other studies consider also the influence of social contacts in the popularity of
media content (see taxonomy in Fig. 2.1 ). For example, the characterization of the
Flickr media collection has been studied in [ 20 ]. An analysis has been performed
along three dimensions: (a) the temporal dimension, which allows tracking the user
interest in a photo over time, (b) the social dimension, aimed at discovering the
social incentives of users in viewing a photo, and (c) the spatial dimension, which
analyzes the geographic distribution of user interest in a photo. Experimental results
reported in [ 20 ] show that users discover new photos within 3 h of their upload.
Furthermore, for the most popular photos (i.e., photo with high view frequency)
almost 45% of the new photo views are generated within the first two days, while
for infrequent images (i.e., photo with low view frequency), this ratio increases to
82%. Moreover, the following two factors affect photo popularity: (a) the social
network behavior of users and (b) photo polling. In fact, people with a large social
network within Flickr have their photos viewed many times, while people with a
poor social network have their images accessed only few times. Finally, the
geographic distribution of user interest is also dependent on the photo popularity.
In fact, the geographic interest in a photo is worldwide when the photo has many
views, while for infrequently viewed photos the geographic distribution is around a
given geographic location.