Information Technology Reference
In-Depth Information
understanding of social streams, online comments, and UGC images and videos will
enable comprehensive tracking to social event progress and evolution. Constructing
association between users' video watching pattern and consuming history will lead
to advanced user modeling and facilitate targeted advertising.
The challenges in exploring social multimedia variety lie in twofold. (1) Orga-
nization. At very high speed, social media generates a torrent of unorganized and
unstructured data. The heterogeneous social multimedia data are unevenly distributed
among various social media networks, and presented as a state in chaotic conglom-
eration. Simple organization on a daily basis or depending on the networks where
the data originate is not efficient enough, in data capture, data analysis, and turning
data into actionable information. (2) Association. Analogous to multimodal analysis,
a key step in exploring the heterogeneous social multimedia data is to establish the
association. However, traditional multimodal analysis assumes that different modali-
ties are associated in semantic-level, e.g., a common semantic subspace exists where
image, audio, and text describing the same concept distribute closely. The association
between heterogeneous social multimedia data is more complex. One single metric
cannot fit to all associations, and the bottom-up strategy fails in most cases.
The user-centric characteristic provides viable solutions to the above challenges.
The fact is that people usually engage in many different social media networks simul-
taneously for different purposes. For example, as illustrated in Fig.
5.1
,thesame
individual may communicate with his/her friends on Facebook, follow real-time hot
events on Twitter, subscribe and watch videos on YouTube, share and discuss favorite
restaurants on Yelp, etc. Anderson Analytics shows that different social media
Fig. 5.1
Social multimedia variety: the overlapped user creates and consumes heterogeneous social
multimedia data
