Using external APIs to improve search | Hacker News Reader

Using external APIs to improve search

gabrielweinberg.com

41 pointsjordanmessina13 comments

@Gabriel: How long do you think this will be a sustainable approach to powering Web search? What I mean is, after a certain amount of traffic will you be able to continue using all these external APIs or do you expect that in the future you will have to augment some or all of this data with something in-house like moving to full-scale Web crawls?

It is quite sustainable. These APIs are central to the business models of the companies and they should generally scale up as needed. For example, SeatGeek already powers WSJ and Yahoo Sports.

Good to know. What are your plans for crawling in the future? Do you think that you will eventually move to crawling as a primary source of result data, and if so, is that time drawing near or do you plan on continuing along the current path for some time?

We never stopped crawling; we just scaled it back for specific purposes, mainly now for spam detection and zero-click info. For the foreseeable future this status quo works well because we can concentrate on our value-ads.

Thanks for that post, I didn't know Qwerly, that might come handy.

I wonder how DuckDuckGo decides which API to query. Gabriel, could you shed some light on this?

Generally we have some kind of entity detection, be it a list or some regex or something more fuzzy. For example, check out the WolframAlpha stuff: http://weinbergalpha.com/

That's great. What's your algorithm for deciding which zero-click info to display when there seem to be multiple options?

For example, I notice that http://duckduckgo.com/?q=fathers+day+2011 doesn't show a WolframAlpha result, but an Amazon result for me.

There is a precedence list that can be tweaked per query, and Amazon is at the bottom, so that is a bug :). In some cases, multiple things are called and the first to answer gets the spot.

Edit: bug fixed.

Interesting, thanks!

One suggestion: I would move the note You can test regexp using DuckDuckGo further to the top. IMHO that's a killer feature!

That page (http://duckduckgo.com/goodies.html) needs a lot of work, generally. I don't have the greatest UI ideas for it yet, but as you might imagine, different things interest different people.

If any HN users want to chat about integrating Qwerly, I'm the founder and available at max@qwerly.com. API is here: http://dev.qwerly.com/.

Gabriel, do you have any stats on proportion of zero click clicks (so probably page views where they don't click any results + clickthroughs from zeroclicks) vs use of standard search?

I do not, sorry.

Reply on news.ycombinator.com