Luis Fernandez-Luque, Randi Karlsen, and Genevieve B. Melton. 2011. HealthTrust: trust-based retrieval of you tube's diabetes channels. In Proceedings of the 20th ACM international conference on Information and knowledge management (CIKM '11), Bettina Berendt, Arjen de Vries, Wenfei Fan, Craig Macdonald, Iadh Ounis, and Ian Ruthven (Eds.). ACM, New York, NY, USA, 1917-1920. DOI=http://dx.doi.org/10.1145/2063576.2063854
Youtube Health Videos: a trust based search approach
1. A
large
number
of
stakeholders
are
publishing
diabetes
videos
Health
Informa8on
is
moving
to
social
media
pla:orms
[1]
(e.g.
blogs,
TwiCer,
YouTube):
Figh%ng
Irrelevant
Health
Videos
in
YouTube:
a
Social
Network
Analysis
Approach
Luis
Fernandez-‐Luque
(luis.luque@norut.no),
Northern
Research
Ins8tute
–
Norut,
Tromsø,
Norway
Randi
Karlsen,
Computer
Science
Department,
University
of
Tromsø,
Norway
Genevieve
B
Melton,
University
of
Minnesota
(Ins8tute
for
Health
Informa8cs),
Minneapolis,
USA
Ignacio
Basagoi8,
ITACA-‐TSB,
Technical
University
of
Valencia,
Spain
Introduc8on
There
are
300+
channels
from
US
hospitals
with
20,000+
videos
(most
of
which
are
on
YouTube).
The
informa8on
society
lives
in
a
constant
state
of
informa8on
overload.
Relevant
videos
Irrelevant
videos
Web
Informa8on
Retrieval
tools
(e.g.
Google,
Bing)
are
widely
used
by
health
informa8on
consumers.
Many
search
algorithms
are
based
on
Social
Network
Analysis
theory.
Hyperlinks
are
used
as
endorsement
indicators.
Hubs
and
PageRank
search
algorithms
are
examples
of
this.
Web
Informa8on
Retrieval
Mo%va%on:
popular
videos,
with
many
incoming
links,
may
not
be
relevant
for
health
informa8on
consumers
(e.g.,
jokes,
polemic
issues,
singers,
spam).
Our
Approach
Two
physicians
evaluated
the
top
20
channels
retrieved
using:
1)
our
approach;
and
2)
YouTube's
search
engine.
Aim:
to
evaluate
whether
or
not
they
would
recommend
each
channel
to
their
pa8ents.
The
agreement
of
the
recommenda8ons
from
both
reviewers
was
evaluated
using
Cohen
Kappa
(0.48-‐moderate
agreement).
Using
YouTube
API;
we
extracted
217
channels
with
the
keyword
“Diabetes”.
5119
videos
and
525
links
were
extracted
(e.g.,
subscrip8ons,
favorites
and
friendships).
We
ranked
the
top
20
channel
authori8es
with
the
HITS
algorithm
[2]
using
JUNG
API
[3].
Methods
Our
list:
Channel
reviewers
recommended
12
out
of
19
(63%)
of
the
channels
from
our
list.
Only
2
out
of
19
(11%)
channels
from
our
list
were
not
recommended
by
either
physician.
YouTube’s
list:
it
had
10
out
of
19
(53%)
channels
recommended
by
both
reviewers.
7
out
of
19
(37%)
channels
were
not
recommended
by
either
clinician.
Note:
two
YouTube
users
removed
their
channels
during
the
evalua8on
process.
Results
By
analyzing
the
YouTube
Diabetes
Community
we
can
infer
knowledge
on
content
quality.
Specially
to
filter
the
less
recommended
channels.
This
provides
promise
for
new
techniques
based
on
Collabora8ve
Filtering
and
Collec8ve
Intelligence.
Our
future
research
will
look
into
a
new
search
engine
for
health
videos
.
Conclusions
Project
Informa8on
Our
approach:
extracted
informa8on
from
the
diabetes
community
in
YouTube
to
find
"high
quality"
channels.
hCp://commons.wikimedia.org/wiki/File:PageRank-‐hi-‐res.png
Favorite
References
[1]
S
Fox,
S
Jones.
The
Social
Life
of
Health
Informa8on.
Pew
Research
Center.
June
2009.
Archived
at:
hCp://www.webcita8on.org/5uSBNoUUr
[2]
Jon
M.
Kleinberg.
1999.
Authorita8ve
sources
in
a
hyperlinked
environment.
J.
ACM
46,
5,
September
1999,
604-‐632.
DOI=10.1145/324133.324140
hCp://doi.acm.org/10.1145/324133.324140
[3]
HITS
Implementa8on
in
JUNG,
Archived
in
hCp://www.webcita8on.org/5uSAOBYNs
This
project
has
been
co-‐funded
by
the
Tromsø
Telemedicine
Laboratory,
a
Centre
for
Research-‐based
Innova8on
supported
by
the
Norwegian
Research
Council.