Post on 16-Jul-2015
CONTEXT AWARE MEDIA PERSONALIZATION
by sentiance with VRT O&I
What TV content would you
recommend for Walter?
It is Tuesday, and Walter just got
home after being stuck in traffic for
more than two hours.
It is getting rather late already, and
Walter has an early day at work
tomorrow. You know he is likely to go
to bed about an hour from now.
The kids are already to bed. His wife
is out. Walter turns on the TV.
ONLY ON WEEKENDS
DOCUMENTARY
Flight
SERIES
Archer
Meta-level or
weighted hybrid
collaborative filtering
incorporates the
Sentiance context
and profile models to
improve upon
traditional user or
content based
recommendations.
Instead of only
learning what Walter
likes, we
simultaneously learn
when he likes it and
why.
Context for
hybrid
recommender
s
KIDS
The Penguins
SERIES
The Goodwife
SERIES
Panorama
COMEDY
The IT CrowdREALITY
The Only
Way is Essex
FILM
Four Lions
SPORTS
Football
Focus
TOO LATE
KIDS ALREADY TO
BED
WIFE’S PREFERENCE
WIFE’S PREFERENCE
MENTALLY TIRED
AFTER EXTREME
LONG COMMUTE
HOME
TOO LONG
TO BED IN 50
MINUTES
EMPATHICPRODUCTSEXPLORATORY RESEARCH TRACK WITH VRT O&I ON MOOD
AND CONTEXT-AWARE MEDIA PERSONALIZATION
• Exploratory
research for small
test group
including
wearables and
activity and
context-aware
apps.
• Larger scale
logging study
providing ‘ground
truth’ on TV
consumption,
moods, context
and need states.
• We observe a
variety of need
states. Each TV
moment can
succeed in fulfilling
certain needs such
as staying up-to-
date, experience
cosy moments,
discovery etc.
• Overall moods /
engagement and
need states are
highly
intercorrelated
need
states &
mood
• Content genres ,
channels, program
titles can be
mapped in order to
investigate what
types of content is
being consumed
for each TV
moment.
• This analysis can
be done for
individuals or
groups of people
content
mapping
• In the same vein,
contextual
variables can be
mapped in order to
investigate how
they correlate with
TV moments.
• In many cases
social company
turns out to be
highly influential
factor.
• Presence or same
room detection is
paramount for
context-aware
recommendations
contextualization
• TV evenings
usually follow
patterns with
fluctuations in
engagement
levels.
• A TV evening can
start as a
background
experience
evolving in a more
engaging
experience. A
evening can be
engagement
rhythm
Need state is a
psychological
concept, hard to
capture in the wild,
however it ‘s a very
powerful concept as
to know what content
fits best with the given
TV moment.
Therefore we need
proxies and develop
technologies and
methodologies to
measure them in
order to include these
proxies in context-
aware recommender
systems.
Mood, social
company, activity
level and other
context variables can
be used as a proxy
for need states.
MOOD, SOCIAL,
ACTIVITY LEVEL
AS PROXIES FOR
NEED STATES
• Investigate what
factors can predict
an increased
preference for
certain TV
moments.
• A long and tiring
day on the road
can change your
content preference
to either equally
engaging but
‘lighter’ content
with less cognitive
load or less
engaging content.
prediction
Anticipate how, when and for how long a user consumes video.
media on the move
You know I am commuting by
train, and that this lasts about
half an hour. Suggest me some
appropriate short-form content.
Once you know why someone
turns on the TV, it becomes much
easier to suggest to them exactly
the content they want to see.
• Engagement is a
hugely important but
no longer the sole
factor as meaningful
TV moments are
experienced at a
variety of
engagement levels,
• SOTA recommender
systems perform
optimally when
engagement is the
goal
• A challenge to
recommend content
when the broader
TV experience
becomes more
important than
content
engagement.
• Social component
recommendations
While our research focused on TV, for
radio too, there is a balance to be found
between engagement and context.
This level of context-awareness is being offered in partnerships with
media recommender system companies worldwide.
WORK WITH US TO
MAKE MEDIA MORE
PERSONAL, RELEVANT &
ENGAGING
info@sentiance.com
THANK YOU FOR WATCHING
CONTEXT-AWARE MEDIA
PERSONALIZATION
BY SENTIANCE
ONLINE WWW.SENTIANCE.COM
CONTACT US AT INFO@SENTIANCE.COM
ALL RIGHTS RESERVED
2015