First, multimedia applications often become
more exploratory, and users are often interested
in images or videos with a particular style,
which is difficult to express and represent. Second,
multimedia data is often used for entertainment
when exploring a visual space, with
no clear end goal.How to discover users’ latent
intent from limited observed data is of paramount
importance in improving multimedia
search and recommendation performance. It
resonates well with the idea that underpins
user-centric multimedia analysis, where the user
profiles, behaviors, and social networks are
sensed, harnessed, and shared to adapt the
results of general multimedia search and recommendation
engines to be more consistent
with user intent.