Feedback Mechanism / Learning for Relationship Extraction

Question

Is there a feedback mechanism that allows us to correct mistakes made by Relationship Extraction so that it becomes smarter overtime and doesn't make the same mistake? eg if it doesn't identify the name of someone as a PERSON or perhaps it is unable to identify something as an ORGANIZATION.

Answer 1

There is no feedback mechanism for Relationship Extraction. If you are trying to recognize a well known PERSON or ORGANIZATION then I would suggest you to use the Alchemy Language API which uses some datasouces for desambiguation.

The picture below shows the entities the subtypes and links to some of the data sources the service uses.

The Alchemy service offers Name Entity Recognition just like Relationship Extraction.

See the demo [here]: http://www.alchemyapi.com/products/demo/alchemylanguage

Answer 2

Javier,

There are two aspects to answering your question.

1) If you are trying to make the Relationship Extraction work better in general (not with regard to a specific example), then we have a new offering coming up that does that. It's called Watson Knowledge Studio, and you can find more information at

http://www.ibm.com/smarterplanet/us/en/ibmwatson/knowledge-studio.html

2) If you are trying to create individual feedback about specific examples. We are definitely considering adding such capability---not to the Relationship Extraction service, but to the Watson Knowledge Studio offering mentioned above, essentially for it to expose an API that collects such feedback and use it in model rebuilding.

Leiming