IBM Bluemix，nl_understanding-文档在哪里

Question

I am trying to gain better understanding how IBM Bluemix natural language understanding works. docs

I found the following example.

import sys
import os
sys.path.append(os.path.join(os.getcwd(),'..'))
import watson_developer_cloud
import watson_developer_cloud.natural_language_understanding.features.v1 as features

nlu = watson_developer_cloud.NaturalLanguageUnderstandingV1(version='2017-02-27',
                                                            username='some_username',
                                                            password='some_password')

nlu.analyze(text='this is my experimental text.  Bruce Banner is the Hulk and Bruce Wayne is BATMAN! Superman fears not Banner, but Wayne.',
            features=[features.Entities(), features.Keywords()])

It generates the following output:

{'entities': [{'count': 3,
   'relevance': 0.915411,
   'text': 'Bruce Banner',
   'type': 'Person'},
  {'count': 1, 'relevance': 0.296395, 'text': 'Wayne', 'type': 'Person'}],
 'keywords': [{'relevance': 0.984789, 'text': 'Bruce Banner'},
  {'relevance': 0.958833, 'text': 'Bruce Wayne'},
  {'relevance': 0.853322, 'text': 'experimental text'},
  {'relevance': 0.627454, 'text': 'Hulk'},
  {'relevance': 0.619956, 'text': 'Superman'},
  {'relevance': 0.583188, 'text': 'BATMAN'}],
 'language': 'en'}

What is relevance in this output? How is it computed? I don't need detailed calculations, as it might be proprietary, but I would like to have basic understanding. I also would like to know how keywords identified? Is there particular corpus used for keywords identification?

Documentation on IBM website is limited.

Answer 1

You can always check out some of the links in our unofficial Watson Landing Page . It has links to the Rebook mentioned above, as well as a link to a nice chatbot implementation that uses NLU.

Answer 2

Relevance: Entity relevance score in the range of 0 - 1. A score of 0 means it is not relevant; 1 means it is highly relevant.

More details in the IBM Redbooks https://www.redbooks.ibm.com/Redbooks.nsf/RedbookAbstracts/sg248398.html?Open