I recently came across a new (stealth) company which is exploring an innovative approach to leveraging NLP in search. The innovation is not just in the matching technology, but also - importantly - in the form that interactions take with the system.
Topodia uses documents as queries, not short strings. It synthesizes a query from the query document and uses that as its starting point. When documents are retrieved for the query, Topodia does further analysis to rank documents by relevance.
The fun doesn't stop there. Interactions with the system are done via a browser plugin, and the resulting collections of documents (which the user can modify) are then shared with the Topodia back end making the collection available to other searchers.
Have a look at their video (original) below and check out their site.
With NLP and semantic technologies getting more visibility, observers generally compare them with the existing paradigm. In evaluating these system, I think it is healthy to look at cases where the front end and UX changes as well as the back end analysis and matching components.
This is known as query-by-example or relevance feedback in the IR literature. It's probably better relevance feedback than exists in many search engines (e.g. "more like this") with a nice interface for organization. But I'm hard pressed to see how this is NLP at all.
Posted by: FD | April 16, 2008 at 12:37 PM
We would like to offer a beta trial to any takers let me know by email.
Posted by: research tools | September 07, 2008 at 11:05 PM
The NLP component operates behind the scenes in topodia. term extraction,using NLP reference sets to deliver quality in meaningful query-by-example or relevance feedback
Posted by: Research Tools | September 07, 2008 at 11:20 PM