Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

One solution may be for Google to radically change their algorithms and policies for web search to de-emphasize phrase-matching and more strongly prioritize inbound links and credibility.

Inbound links and the calculated "credibility" from the same are what killed the web the first time around. There was once a democratized web era when that actually worked -- when millions of people had their little Goecities pages and were linking the cool stuff -- but in the modern era it's 99% consumers who cast no votes, and the last 1% is extraordinarily incestuous circular link love: Marcos links to Coding Horror who links to Daring Fireball who links to Scoble who links to Marcos, etc.

People with neither information or authority end up being the credible authority on matters they have aren't authorities on. Scoble a few years back pointed out the fact that according to search engines he was the most important Robert in the world. That is a frightening concept.

We will move from an era of search engines to an era of expert engines. Many of the questions I used to "ask" Google I now ask of Wolfram Alpha, and its approach has turned out to be quite useful. Expand that computer knowledge more broadly, and improve the human syntax parsing. and we'll have a winner. Several such systems are built around computer learning of the wikipedia corpus.



I noticed recently that Google has started to give explicit answers to some search queries - e.g http://www.google.co.uk/search?q=json+content+type


Yup, that is exactly that sort of "machine knowledge" that is beneficial.

I suspect where their algorithm isn't somewhat sure of the answer it essentially does A:B testing with users -- a while back I asked it who the governor of Illinois was and it replied Blagojevich (long after he was ousted). I pointed it out to a friend and they tried and got Quinn. In both cases it gives you the option of flagging whether it is wrong, though it seems suspect to let people who are asking the question in the first place declare its rightness, beyond egregiously wrong answers.


Several such systems are built around computer learning of the wikipedia corpus.

Any ones in particular that you've found work well?


Powerset worked well enough that Microsoft bought them for $100 million.


Cuil, of course!




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: