Friday, February 12, 2010

The broken tail of queries

I made a huge mistake in the last post...
It's not a long tail of results, but it's a long tail of queries.
And in fact, it's not even a healthy, smooth long tail, but a broken long tail.

Let me explain myself.

Let's get back first to Wikipedia.

"The Long Tail or long tail is a retailing concept describing the
niche strategy of selling a large number of unique items in relatively
small quantities – usually in addition to selling fewer popular items
in large quantities."
Let's look at the graph that illustrates that concept.

It's just a graph showing popularity ranking. To the right is the long
tail; to the left are the few that dominate. Notice that the volumes
of both areas match.

So I made a second mistake.
It's not 20-80 but 20-50. In other words, 20% of all items account for
50% of all downloads. The remaining 80% account for the remaining
That's less that previously thought. In the previous post, it was
believed that those 20% of total items accounted for 80% of total

Now, let's get back to the search world and use "results" and
"queries" for metrics.
Since popular queries generate lots of results and specific queries
generate few results, we can say that we get a graph that shows a popularity ranking.
In fact, we should get a long tail of queries:

But as noticed in the previous post, we have a problem: current search
engines are only able to handle very simple queries, those queries
comprised in average of 2.1, boring keywords.
Thus, most of the queries give no results.
In fact, we get eventually a broken tail of queries:

That's a pity. That means that 50% of the time we get nothing.

You would tell me that's not exactly what's happening.
Because there are cases where users know exactly the name of the
product they are looking for.
In that case, they are using the search bar as a fetch dog and if the
item is in the database, well indeed that will yield results.
But when you wanna discover new things based on personal criteria, do
you know the exact name of the item you are looking for?
The answer is of course no.
So it doesn't change drastically the previously tail of queries, perhaps delay a little bit the moment when the graph breaks down.

Once the user has obtained the results, he chooses the items that he
likes and then can decide whether or not to download them.

Since a download automatically translates into a sale (assuming that
download are not for free), we get eventually a huge zone where the
sales that should have been done have vanished:

That's a huge loss of money for the retailer and a huge loss of time
for the user.

But ascot project has come up with a new way to specify the search
criteria and a new way to deal with those specific, customized, personal queries.
And we are confident that we are able to give good results where
current search engines are failing to give any results.

In fact, that's a win-win situation for both users and retailers:
users will be able to discover new items and retailers will be able to
make cash on items that were not reached so far...

No comments:

Post a Comment