Oleg Zabluda's blog
Friday, May 18, 2012
 
If computers had any kind of "understanding", a lot of things would be better, like spell/grammar checking, machine...
If computers had any kind of "understanding", a lot of things would be better, like spell/grammar checking, machine translation, speech recognition, etc... The problem is that we don't really know how to represent "knowledge", and not for the lack or trying. One of the easiest thing to start with, is an entity-relationship graph (http://en.wikipedia.org/wiki/Entity-relationship_model).

The very recent history of it is Freebase with 22M entities and 80M relationships (http://en.wikipedia.org/wiki/Freebase), made by company called Metaweb (http://en.wikipedia.org/wiki/Metaweb), founded in July 2005, and acquired by Google in July 2010. Now, 2 years later, Google released its "Knowledge Graph", with 500 M entities, and 3.5 G relationships, which may be a major development in this area, initially applied to search.

http://googleblog.blogspot.com/2012/05/introducing-knowledge-graph-things-not.html

I have to do more research on how they collect entities and relationships, but some of recent things about it are here:

http://en.wikipedia.org/wiki/Microformat (not widely used, maybe because before the Knowledge Graph there wasn't enough incentive)
http://en.wikipedia.org/wiki/Wikidata (partially funded by Google)

See also:
http://en.wikipedia.org/wiki/AI_winter
http://en.wikipedia.org/wiki/History_of_machine_translation
http://schema.org/
http://googleblog.blogspot.com/2012/05/introducing-knowledge-graph-things-not.html

Labels:


- posted by Oleg Zabluda @ 4:07 PM | permalink | | | technorati | bloglines | feedster
| |

Home

Powered by Blogger