Ben Gomes,
engineer at Google, explains how the search engine really works
We all use
Google search but have you ever wondered how it really works. How does Google
manage to pick relevant data from the mountain of information available on the
Internet within a matter of seconds?
Business Line spoke to Mr
Ben Gomes, a distinguished engineer at Google's headquarters in Mountain View,
where he is a lead for the company's efforts on search features.
Mr Ben, who did
his schooling in Bangalore, says that making changes to a search engine which
gets billions of queries a day is kind of like changing out the engine of a
jumbo jet while flying at 30,000 feet. Google does this through carefully
designed experiments to listen to millions of users.
In fact at any
given time, Google might be running up to 200 live experiments and you may have
been a guinea pig even without knowing. Excerpts from the conversation.
How has search
evolved?
When I first
started working in Google some searches could take up to 20 seconds and today
it would be shocking if it took more than a second. It's become faster and
faster. It's even more astonishing when you look at the amount of data that's
grown many folds. But more remarkably search has gone far more accurate than
when we had less amount data to search through. We put huge amount of effort in
getting newer algorithm that understands language, documents and matching what
users want.
In Google, we
get over a billion queries a day so we have 100 of millions of users who can
give us information on whether what we are doing is good or not and that too
extremely quickly. We take advantage of this for testing of ideas. What we have
is a factory of new ideas. We ran over 58,000 experiments from this factory of
innovation listening to millions of users.
We launch about
500 changes every year. These changes are like a complex machine such as a jet
engine and what we have to do is to change the engine as you fly to make it
faster, efficient and more comfortable to users.
How do you
carry out such complex changes?
We have built
infrastructure over the years that does this. At any given time, we would be
running a couple of 100 experiments with users who are online. After an idea
has been tested, senior engineers look at the implementation and kind of
complexities. If it passes muster then it goes to launch. So an idea goes
through a rigorous process before it's launched
When a user
puts in a search query how does Google know which links to show first?
There are many
factors. When Google started, the words of the query mattered a lot. So we used
to find pages that matched the words exactly. Over time we have got very
sophisticated in how we understand your query. We process the query you type
and understand all the meanings and synonyms. If you used the word ‘change' it
means ‘adjust' but if you type ‘change your tires' we know it doesn't mean
‘adjust your tyre'. Then we send that to our computers that look at the index
of data of the Web using semantics of your query. But that's not enough. We
need to get you data from a good source that you would trust. The degree of freshness
in the information also matters so we look for the latest information on your
query. The next thing that we take into account is your location. If you are in
Delhi and you are looking for a restaurant there's no use if I show you info
about a restaurant outside the city. These are the factors that get into making
search relevant for you. We have data centres all over the world and on an
average your query travels over 1,500 miles.
How do you deal
with those who claim to have broken Google's algorithm to ensure that their Web
sites show up first on search pages?
We have
developed tools to battle this. There's a lot of money in this for those who do
this kind of activity but we are many steps ahead of them. It's a constant
battle against people trying to fool us. This is a key area where we have
invested.
Some claim that
Google throws up links based on the advertisement it gets.
We have strong
dividing line between search and advertisement since very early days. We have
to get the right search results irrespective of the revenue. We model the way
how newspapers divide editorial and advertising. We want to be making the
product relevant for the users, monetising will follow.
Where is search
heading? Do you think Apple has taken the next step with its Siri?
Next step for
Google is towards knowledge. By that what we want to do is how we can provide
answer to your questions. For example, if you want to know how tall is empire
state building then I should be able to give that answer.
Beyond that we
would want to give you information what you should be getting. For example, you
think about fast food and there is a lot of raw information about calories but
we would like to tell you eat a salad instead. We want to give you knowledge
than just raw data. We are on that path although just digging at the surface.
No comments:
Post a Comment