The biggest problem Google has, in my opinion, is it's identity crisis. It's becoming a stranger.
Googles popularity came from it's simplicity. The services Google started buying (looking at YouTube here) were gaining traction because they were the simple equivalent to their competitors.
Today? Not so much. YouTube has gradually become more alienated since 2009. Yesterday I was watching a video, and not a single video listed in the sidebar (once called "Related Videos") was actually related to the video I was watching. The uploader was a popular YouTuber with easily over a hundred videos (No idea how much exactly because I couldn't find that number) yet all the suggestions were music videos from my usual browsing.
Context switching is no more, because everything is being overengineered to keep you in your own content bubble. Yet, that stupid auto generated playlist in the sidebar that I've never clicked, keeps lurking there on every video, for days at a time, before changing into another playlist that is no more appealing to me than the previous one.
If all the effort of that huge datapool we are selling our souls to is to make advertisers happy, and the users don't get anything out of that effort but more abstraction and more generalized data science slapped onto a new UI every couple years, it's not a fair trade. And people won't put up with that forever.
Google is still going strong. The foundation that put it in its place, is crumbling. A big chunk of their business still relies on that, though, so I think these articles have a fair point.
> all the suggestions were music videos from my usual browsing.
Good observation! Google blew it!
When I go to YouTube, I don't see what you saw,
but then I don't
'log in' to Google or accept cookies from them.
So, the videos I see on the right are related
to what I am watching at the time instead of whatever
I've watched, searched for, etc. in the past.
I can believe that Google is doing what you
describe, and this is a symptom of totally
wacko data science and brain-dead recommendation
engine construction. Where from, why?
One approach to recommendation is to try to
say what a given user likes. So, look at
all their activity, say, products they've looked
at at Amazon, videos they've seen at YouTube,
searches they've done at Google, Web sites
they've visited as determined by following
third party cookies, etc. Then my view
is:
(1) For ad targeting, in the short term
(that is, when displaying an ad
only a short time after the data used
for the targeting)
maybe okay for effective ad targeting,
assuming it doesn't
offend the user.
(2) For content, nope, won't work and
with your experience a solid example
of why not.
Why? Here is a hypothetical example:
I go online and search for
flowers and chocolate candy and
have them delivered gift wrapped;
similarly for some things at
Victoria's Secret.
So, from then on I get recommendations
for flowers, candy, anything chocolate,
and women's frilly undies.
Ha! I'm
a fully normal, heterosexual male
and don't much care for distaff stuff!
So, why'd I buy the flowers? Sure:
As Valentines gift for my wife, once
a year! The other 364 days of the
year, f'get about it!
Or, I shop for some DVDs of some
old Disney movies. Does this mean
that I like old Disney movies instead
of, say, movies about Tom Clancy stories?
Nope! Instead I was just shopping for
some DVDs to entertain the children
of some of my friends my wife and
I had over for a nice
BBQ and beer on the back porch.
Or, as I see it, for something better,
what a person likes at a given
time should to be for some one of their
interests at that time. Then the
recommendation engine has to
learn about that interest at that
time.
A biggie is that that interest
is likely some narrow thing, narrow
in time, circumstances, etc. So,
past browsing history, shopping,
watching, etc. should be treated as,
first cut, irrelevant or, in probabilistic
terms, independent of what the heck the
person wants in their present context.
BTW, with mild assumptions,
probabilistically independent implies
(statistically) uncorrelated, although
in the usual treatments independence is
much, much more general, say, is in terms
of sigma algebras generated by some
sets, possibly uncountably infinite,
of random variables, and such a definition
for uncorrelated is rarely or never given.
With high irony, likely search results
at Google from the keywords/phrases
someone enters are likely independent
or nearly so of anything else Google
knows about the person. Or, if at Google
search type in
"I'm shocked, shocked to learn that
gambling is going on here"
then should get back the script of the
classic movie Casablanca and
don't expect to get back
results about flowers, chocolates,
flimsy undies, and Disney movies,
or Tom Clancy movies either.
It's possible to use butter, milk, eggs,
flour, Kirschwasser sugar syrup,
cheeries,
chocolate, etc. to make a
fantastic cake or a really big mess.
Same for using data science.
Watch here on HN when I announce my
recommendation engine (soon,
currently mud wrestling with DVD
burners)
that will
treat each user's interest as
unique in all the world,
have
the best protections of user privacy,
and do nothing with and have nothing
on anything about the user before
they requested their recommendation.
When the recommendations come back,
the ad targeting may have to do with
just those recommendations but certainly
not with some shopping for flimsy
undies a week before Valentine's day.
Google's search engine is just terrific
for a lot of the content on the
Internet, and where Google is good
my work will not be better.
But as your experience illustrates,
for some searches there is room
for something better. My search
engine has nothing to do with
keywords/phrases; my view is that
what I've developed stands to be
much better for a significant fraction
of the content on the Internet,
searches people want to do,
and results they want to find.
But, again, for where Google works
well, and sometimes it is fantastic,
my work is not better.
Googles popularity came from it's simplicity. The services Google started buying (looking at YouTube here) were gaining traction because they were the simple equivalent to their competitors.
Today? Not so much. YouTube has gradually become more alienated since 2009. Yesterday I was watching a video, and not a single video listed in the sidebar (once called "Related Videos") was actually related to the video I was watching. The uploader was a popular YouTuber with easily over a hundred videos (No idea how much exactly because I couldn't find that number) yet all the suggestions were music videos from my usual browsing.
Context switching is no more, because everything is being overengineered to keep you in your own content bubble. Yet, that stupid auto generated playlist in the sidebar that I've never clicked, keeps lurking there on every video, for days at a time, before changing into another playlist that is no more appealing to me than the previous one.
If all the effort of that huge datapool we are selling our souls to is to make advertisers happy, and the users don't get anything out of that effort but more abstraction and more generalized data science slapped onto a new UI every couple years, it's not a fair trade. And people won't put up with that forever.
Google is still going strong. The foundation that put it in its place, is crumbling. A big chunk of their business still relies on that, though, so I think these articles have a fair point.