Extractiv’s novelty is that we run our NLP on a web crawler, so you can transform the open unstructured web into structured semantic output. This is great when you want to discover information from the web as opposed to process collections of documents you already have. For the latter case, we provide a REST-based on-demand service.

We currently provide over 150 entity types, relations, and more, in output formats including JSON and RDF. I’m with Tom Tague when he said “Entities describe the news – facts and events are the news.” We like entities, but want to provide a lot more than that. We’re new, so bear with us as we still have some additional services to roll out, including entity linking, which should be within a few weeks.

To usher us in as a late contender to this post, I’ve put URLs by which you can get live results on these 3 transcripts (or any URL), which hits our demo server. I promise I didn’t doctor the results. :)

http://rest.extractiv.com/extractiv/?url=http://rest.extractiv.com/sample_docs/viewchange1.txt&output_format=html_viewer#
http://rest.extractiv.com/extractiv/?url=http://rest.extractiv.com/sample_docs/viewchange2.txt&output_format=html_viewer#
http://rest.extractiv.com/extractiv/?url=http://rest.extractiv.com/sample_docs/viewchange3.txt&output_format=html_viewer#

website: http://www.extractiv.com
blog: http://blog.extractiv.com
docs: http://wiki.extractiv.com
twitter: http://twitter.com/#!/Extractiv (@extractiv)

@tomayac

Curious which NLP APIs from Alchemy were used in this comparison: AlchemyAPI provides both a Zementa-style “Concept Tagging” API and a traditional “Entity Extraction” API.

If you’re receiving a low number of entity results, you’re likely using only one of our APIs. For the maximum number of semantically linked results (for Keywords/Tags AND named Entities) be sure to use both Named Entity Extraction ( http://www.alchemyapi.com/api/entity/ ) and Concept Tagging ( http://www.alchemyapi.com/api/concept/ ).

AlchemyAPI’s Concept Tagging API returns disambiguated and linked-data-enhanced results for a wide variety of concepts and named entity types. When comparing against solutions such as Zementa you’ll find this API provides a more “apples to apples” comparison.

http://github.com/robdiciuccio/Simple-API-Test-Tool

Just to mention, Kosmix.com builds topic pages for any entity and might be worth mentioning as a content provider. For example, http://www.kosmix.com/topic/akon or http://www.kosmix.com/topic/Splice_(film)

Also, Kosmix provides an api to recognize, disambiguate, and link entities in a document. For example, http://www.healthboards.com/boards/showthread.php?t=748185

Just to mention, I work at Kosmix, and appreciate your article comparing different products.

@Bob McCreary:
Zemanta isn’t designed to run on such short sentences (albeit it works most of the time). Try it with larger article. Also you seem to be looking at “keywords” instead of “markup” part of the response.

bye
Andraz Tori, CTO at Zemanta

Zemanta doesn’t get “John Smith”, or “Paris”.

Instead it comes back with “Georgia” (strange), “United States”, and “Lamar County”.

Many of the other listed APIs are detecting the entities missed by Zemanta, so I’m at a loss as to why it’s listed first. Seeing similar results for pretty much every news article I try.

Another recent look into this space by Michael Fagan of the Bing Maps team listed OpenCalais near the bottom of the rankings, yet it’s listed as #2 here.

Here’s a link to fagan’s review:

http://faganm.com/blog/2010/01/02/1009/