►
Description
Dexter Lee and John Joyce from Acryl Data reveal recommendations in the new DataHub landing page experience - making it easier than ever for DataHub users to navigate their metadata.
See Recommendations in action in the DataHub demo site: https://demo.datahubproject.io/
Join us at our next Town Hall - RSVP here: https://forms.gle/g8EpCLnohtPLLtdg6
A
Yeah,
so
we're
gonna
talk
a
little
bit
about
a
new
feature,
our
new
subsystem
that
the
indexer
have
been
working
on
in
the
past
month
and
we're
gonna
actually
just
go
right
into
a
demo
right
off
the
bat
and
for
I
think,
the
first
time
ever
we're
going
to
be
demoing
on
the
demo
site
directly
because
it's
already
released
in
master.
A
You
know
some
of
the
categories
here
you
might
be
familiar
with.
This
is
basically
just
a
rebranding
of
what
we
had
here
before,
with
a
little
bit
more
information,
as
you
can
see.
But
really
the
interesting
part
is
when
you
scroll
down
a
bit
and
what
you'll
see
is
a
series
of
modules
and
each
one
of
these
modules
represents
a
group
of
recommendations.
A
So
you
can
see.
First,
we
have
all
of
the
platforms
that
have
been
ingested
into
your
data
hub
instance.
You
can
go
and
click
on
them
and
actually
jump
right
into
search,
but
you
can
also
see
we
have
a
couple
other
categories
right
off
the
bat
here
recently
viewed.
So
this
is
things
that
I,
as
a
user,
have
recently
viewed
profiles.
You
can
also
see
most
popular
entities
on
datahub.
A
So
what
we've
done
is
we've
tried
to
build
this
subsystem
for
pushing
server
driven
recommendations
in
a
way
that
can
be
kind
of
reused
across
multiple
surface
areas
within
the
application.
So
what
you're
seeing
is
sort
of
the
home
page
scenario,
but
if
you
go
ahead
and
search
something
let's
say:
sample
you'll
actually
find
that
we
surface
similar
recommendations
on
the
search
page
as
well,
and
maybe
not
as
useful
in
the
case
that
you
have
a
lot
of
relevant
results.
A
But
in
the
case
that
you're
missing
something
it
can
be
pretty
useful
as
a
jump
off
point
to
kind
of
explore
and
find
additional
metadata
that
you
may
not
have
even
known
existed
in
your
instance.
So
it's
pretty
exciting
I'll,
also
plug
a
new
button.
We
put
here
called
explore
your
metadata,
which
will
just
link
you
to
basically
a
view
of
all
of
the
metadata
in
your
instance,
which
I
think
is
another
way
to
kind
of
encourage
exploration
across
datahub.
A
A
This
is
the
third
surface
area
that
we
can
attach
recommendations
to
from
the
server,
so
there's
kind
of
three
broad
surface
areas
today,
the
home
page,
the
search
results
page
and
then
additionally,
the
the
entity
profiles,
and
now
I'm
going
to
talk
a
little
bit
about
sort
of
the
motivation
behind
this
recommendation.
Subsystem
and
I'll
pass
it
over
to
dexter
to
talk
about
how
we
built
it.
A
So
really,
what
we
set
out
to
do
is
to
kind
of
make
it
easier
to
find
metadata
that
you
care
about
and
metadata.
That
is
just
in
your
instance
already,
if
you
guys
recall,
we
just
had
those
basic
categorizations
on
the
home
page
previously
and
they
weren't
very
useful
to
understanding
what
data
you
even
have
in
data
hub
you'd
have
to
kind
of
manually
click
around
and
investigate
quite
a
bit
on
your
own.
Here.
A
We're
actually
trying
to
take
all
of
your
metadata
and
and
draw
some
insights
and
push
it
to
you
from
the
get
go
so
there's
a
couple
things
we
were
looking
to
do
here:
improve
the
exploration
process
by
providing
kind
of
guided
navigation
to
sort
of
that
high
value,
metadata
and
you're.
Seeing
that
with
the
most
viewed
categorizations
that
we've
added
you
know,
we
want
you
to
get
to
the
metadata
that
you
care
about
in
less
clicks.
What
I
touched
on
earlier.
A
We
wanted
to
be
able
to
provide
sort
of
high
level
insights
about
your
metadata
and
push
those
to
you
and
then
finally,
we
wanted
to
add
sort
of
personalization
or
have
a
place
to
add
personalization
suggestions
about
metadata
that
are
kind
of
relevant
on
a
per
user
basis,
and
so
what
we
did
is
kind
of
try
to
take
all
of
these
requirements
and
build
one
subsystem.
That
would
help
us
to
address
all
of
these
verticals
with
one
go,
and
I'm
gonna
pass
it
over
to
to
dexter
to
talk
about
how
we
did
that
awesome.
B
So
the
goal
that
we
set
out
for
was
to
create
an
extensible
framework
for
generating
these
recommendations.
So
we
don't
want
extra
work,
extra
work
on
the
back-end
side,
extra
work
on
the
front-end
side,
as
we
try
to
introduce
more
recommendations
to
to
datahub
so
number
one.
We
try
to
make
sure
we
can
share
code
between
multiple
surface
areas,
so
we
already
showed
three
places.
B
The
second
is
to
create
a
modular
server
driven
architecture,
because,
through
my
previous
life,
I've
seen
a
lot
of
pain
where
we
started
in
the
front
end,
and
then
we
had
to
move
everything
to
the
back
end
because
to
run
good
data
analysis
and
experimentation,
it
was
always
much
more
useful
to
have
a
server
driven
architecture
for
recommendations
rather
than
the
front
end.
B
So
what
we
tried
to
do
was
decouple
front
end
from
the
ui
all
right
front
end
from
the
back
end,
so
the
back
end
has
its
own
ways
of
generating
these
recommendations,
but
how
they're
rendered
in
the
front
end
is
in
a
different
way,
so
we
kind
of
categorize
these
things
into
a
few
small
render
types
like
the
entity
list
or
the
tag
list,
platform
list
and
so
on,
and
then
in
the
ui
for
each
of
these
render
type.
B
We
have
additional
code
to
render
these
so,
let's
say
because
we
already
set
up
for
entity
list.
If
we
have
different
algorithms
for
generating
a
list
of
entities
to
recommend,
we
don't
have
to
change
front
end.
We
only
need
to
change,
backend
and
add
a
new
candidate
source
to
our
system
that
the
last
one
is
to
support
some
out-of-box
support.
We
have
added
top
platforms,
most
viewed
recently
viewed
top
tags
and
top
glossary
terms
as
some
examples
of
what
we
could
add
to
on
to
our
system.
B
Let's
move
to
the
next
one.
So
how
do
we
do
it?
So
first
the
ui
sends
the
server
request
for
recommendations
this.
In
this
request,
I
need
to
send
enough
context
for
the
server
to
know
where
these
recommendations
are
being
surfaced.
So
number
one
is
the
user?
That's
recommending
it
so
that
we
can
have
some
personalized
recommendations,
but
also
the
scenario
type.
That
is
what
we
call
where
it's
trying
to
get
surface
so
home,
page
search,
page
or
entity,
page
and
so
on
and
so
forth.
B
We
also
try
to
take
in
as
much
context
as
possible,
so,
for
example,
in
a
search
page,
if
we
know
the
query
that
the
user
was
searching
for,
we
can
perhaps
send
better
recommendations
in
the
future,
so
we
try
to
get
additional
context
there.
Now
once
it
comes
to
the
server,
it
goes
through
multiple
candidate
sources
that
we
have
defined,
which
each
of
them
have
eligibility
based
on
the
input
scenario,
type
and
they're
all
sourced
in
parallel
and
there's
a
single
ranker
that
ranks
them
and
chooses
which
modules
to
send
back
to
the
ui.
B
Now
these
modules,
each
of
them
contain
content,
so
each
element
inside
the
module,
as
well
as
a
render
type.
B
That
is,
that
is
what
we
just
explained,
where
each
render
type
here
in
the
ui
corresponds
to
a
different
rendering
logic
in
the
react
side,
so
each
different
render
type
corresponds
to
a
different
way
of
showing
it
in
the
ui
now
so
once
it
goes
into
the
react,
the
renderer
passes
it
on
to
the
correct
rendering
logic
based
on
the
render
type
and
then
finally,
it
is
displayed
on
the
left,
where
you
have
the
module
title
and
the
content
shown
as
well.
B
Now
each
content
has
some
landing
page
information,
so
it
it
has
information
about.
If
you
click
on
it,
where
would
you
land,
because
sometimes
you
want
to
land
it
on
the
entity
page?
Sometimes
you
want
to
land
it
on
the
search
page
and
so
on
awesome.
So
this
is
check
it
out,
check
it
out,
live
in
demo
and
give
us
feedback
or
any
ideas
about
additional
recommendations.
We
should
support
out
of
the
box
and
so
on.
Thank
you.
B
A
Yeah,
I
just
want
to
conclude
by
saying:
we've
tried
to
deliberately
make
this
super
super
modular
so
that
we
can
source
recommendations
from
the
community.
So
if
you
guys
have
interesting
ideas
around
how
to
add
more
insights
recommendations,
any
part
of
the
app
please
do,
let
us
know
we're
definitely
taking
feedback.
What
you
see
on
the
app
now
is
kind
of
the
first
pass
of
recommendations,
and
hopefully
they
will
be
useful,
as
is
but
I'm
sure,
there's
a
lot
more
that
we
can
do
to
improve
it
in
the
future.