How do I create a new corpus for Concept Insights?
When I run:
curl -D - -X POST -u myuuid:mypass -d '{"access":"private","users":[{"uid":"myuuid","permission":"AdminReadWrite"}]}' \ 'https://gateway.watsonplatform.net/concept-insights-beta/api/v1/corpus/myuuid/corpname'
I get this error:
{"error":"corpus does not exist","url":"/myuuid/corpname"}
Any ideas?
(X-Global-Transaction-ID: 82095273)
Nevermind...
There's a typo in the docs. I changed "POST" to "PUT" and it worked!
Awesome!
you should mark your own answer as accepted :-P
Thanks for finding the error. We'll change that example.
@mfrager, have you had success adding documents to a corpus in Concepts Insights? Typically you will want the documents to have enough text so that at least 10 concepts from Wikipedia are successfully detected by Concept Insights. Also in our experience it starts to show useful behavior when the number of documents added is in the hundreds or a few thousands.
I was able to upload my corpus. I'm pretty happy with how it all turned out and I plan on continuing to work with the Concept Insights API, I think it has great potential.
If you'd like to take a look at the project I'm working on here's some details. This is all in alpha development of course...
I've loaded up the entire congressional record from 2014 and linked it to a page for each representative. Then I'm using the Concept Insights API to do a topic-based search on the record transcripts.
Here's the link: http://atellix.com/search
A few tips, you'll need to select "Congress Records" from the search, and then enter a partial topic and then click on the topic you want from the drop-down menu to search before pressing return (otherwise it won't work).
For an example try doing a search for "Immigration" or try this link for a short cut:
http://www.atellix.com/search?search_mode=congrec⊤ic=Immigration
Once you search you'll get a list of the congressional transcripts that contain a discussion related to the topic, the members who participated in those discussion, and a list of related topics.
Your use case seems extremely interesting, is it still online ? I can't access the website
Awesome, we tried it out. It's good, it feels that it has a clear use case.
Quick suggestion. When you do a semanticSearch, you also get back a list of concepts with the coordinates within which they appear in each of the documents. There is a "score" field that tells you how relevant the concept (within the document) is to the query. You could use this data in order to do some highlighting of the documents when you click them.