►
From YouTube: Predictive Web Browser: 2014 Spring NuPIC Hackathon Demo
Description
Julie Pitt. Source code: https://github.com/oomagnitude/duke-of-url
A
Can
you
guys
hear
me?
Okay,
all
right,
excellent?
My
name
is
julie
and
I'm
going
to
talk
about
a
concept
for
a
predictive
web
browser.
My
inspiration
for
this
is
that
you
know
as
a
power
web
user.
I
often
drill
down
into
obscure
pages.
For
example,
if
I
go
to
flickr,
I
want
to
download
an
image.
There's
like
three
clicks.
A
A
You
have
to
make
a
lot
of
decisions
when
you're
developing,
say
a
consumer
product
with
the
ui.
You
have
to
decide
what
button
should
go
where
which
one
should
we
hide?
Which
ones
should
we
put
up
front
and
that
generally
works
for
them?
You
know
most
users,
but
what?
If
there
was
an
intelligent
ui
that
would
learn
your
behavior
and
actually
surface
the
most
interesting
features
or
options,
so
this
is
really
just
baby
step.
This
is
the
first
time
I've
ever
gotten
my
own
data
working
in
new
pic.
A
So
all
I
really
started
with
was
my
own
browsing
history
off
of
chrome.
I
just
basically
took
it
off
of
the
sqlite
database
and
came
up
with
basically
a
parsed
url
format
that
takes
each
chunk
and
feeds
it
into
the
cla
as
a
each
as
a
field,
and
so
my
web
browser
is
not
terribly
beautiful.
Yet
it's
basically
what
I
did
was
I
took
that
browsing
history
over
the
last
90
days,
I
was
a
little
bit
shocked
to
find
out
that
I
had
18
000
page
visits
in
the
last
90
days.
A
So,
probably,
first
and
foremost,
the
most
likely
thing
I'm
going
to
hit
is
google.com
and
I
basically
did
a
multi-step
prediction
where
it
would
say
what
would
be
the
next
click
and
what
would
be
three
clicks
out
and
not
surprisingly,
it
thinks
that
I'm
going
to
stay
on
google,
and
I
guess
I
like
linkedin
too
so
by
the
way.
This
is
looking
at
the
first
six
path
elements,
so
there's
kind
of
some
limit
to
how
much
of
the
url
it
can
actually
look
at.
A
A
Okay,
but
but
github
I
go
to
github
a
lot
right.
This
is
actually
the
url
to
the
code
for
this
project
and
google
again,
but
there's
a
30
chance
that
I'm
going
to
go
to
github
next.
But
what
if
I
go
to
one
of
the
new
menta
source
codes,
what
will
it
show
me?
It
will
actually
predict
that
I'm
going
to
stay
on
github,
so
probably
what's
happening.
There
is.
I
have
my
own
personal
project
page
up
a
lot,
and
I
use
that
as
a
launching
pad
versus.
A
So
you
kind
of
get
the
idea.
I
could
probably
keep
getting
more
examples,
but
that's
kind
of
the
first
step
along
the
way.
I
think
my
biggest
challenge
here
is
going
to
be
having
enough
data
to
train
this
on,
because
I
don't
expect
the
average
person
to
send
their
browsing
history
to
me
over
the
internet.
But
if
anybody
knows
of
any
data
sets
to
try,
that
would
be
great.
B
So
this
is
really
cool.
I
think
this
is
another
one
of
those
things
that
could
be.
The
core
of
this
could
be
used
in
so
many
different
ways.
You
know
you
mentioned
a
couple
of
them,
but
just
being
able
to
do
click
prediction
of
websites
is
like
one
of
those
holy
grail
problems
that
everyone
is
trying
to
solve,
and
the
fact
that
you
can
do
it
and
showing
the
multiple
predictions
at
each
time
step
and
as
well.
Multiple
steps
into
the
future
is
really
cool.
C
C
A
C
D
D
It
well,
that
would
be
the
idea,
that's
something
we
could
easily
work
on,
but
I
might
understood
this
so
it
would
basically
say:
okay,
if
you
had
it
wasn't
possible
to
do
more
than
that
at
this
point
in
time
it
says
well,
most
likely.
The
next
thing
you're
going
to
click
on
is
something
with
the
domain
of
github
right.
So.