winkjs / wink-ner Goto Github PK

var ner = require( 'wink-ner' );
var myNER = ner();
myNER .defineConfig({ignoreDiacritics: true, tagsToIgnore: ['punctuation'], valuesToIgnore: []});

var trainingData = [
  { text: 'T-Mobile', entityType: 'company', symbol: 'TMUS' }
];

myNER.learn( trainingData );
var winkTokenizer = require( 'wink-tokenizer' );
var tokenize = winkTokenizer().tokenize;
var tokens = tokenize( 'I work for T-Mobile and I like it' );
tokens = myNER.recognize( tokens );
console.log( tokens );

'T-Mobile' is not recognized although punctuation is specified in defineConfig.

Examples are:

1. UK or U.K. or U K
2. Kg or K.g.
3. IBM or I.B.M. or I. B. M.

For example, if I want to identify a price, is there a way to define a learning entry composed of a currency token followed by a number token?

Hello,
I play around with the library. My training set leads to a lot of wrong entityType assignments, but I want to test it a bit more.
With my trained ner model. Is it possible to use that with wink-nlp? I haven't seen any docs for that. I would like to use the visualization features showcased in the wink-nlp with a custom annoted ner set.

Hello, nice package! Do you have a default training data file, so that I could train the model on some default settings to get it up and running? Rather than having to create the training data file myself? Thanks!

1. mocha

The examples shows how NER can identify and distinguish between Manchester (the city) vs Manchester United (the club).

However, I can't find a way to do the opposite.

Consider USA and United States of America. Both should be detected as the same entity.

Is this possible?