nalapa/modules/BIOLabel.js

Collection of NodeJS NLP Library for Bahasa Indonesia.

/*! Nalapa v1.10.0 - MIT license */

'use strict'

/**
 * Module dependencies
 */
var _ = require('lodash')
var Tokenizer = require('../modules/tokenizer.js')

/**
 * @param {}
 * @return {}
 * @api public
 */

var BIOLabel = function () {}

BIOLabel.prototype.label = function(_data) {

  var text = (_.isUndefined(_data.text)) ? '' : _data.text
  var tokens = Tokenizer.tokenize(text)

  var data = {}
  data['tokens'] = tokens
  data['labels'] = new Array(tokens.length)
  for (var i = 0; i < data['labels'].length; i++)
    data['labels'][i] = []

  var biolabels = _data.labels
  biolabels = (_.isUndefined(biolabels)) ? [] : biolabels
  
  biolabels.forEach(function (biolabel) {
    var label = biolabel.label
    var words = biolabel.words

    words.forEach(function (word) {
      var tokens = Tokenizer.tokenize(word)

      // get array of match val between text and  label word (true|false)
      var matchArray = _.map(data['tokens'], function (edata, idata) {
        var matchLabelWords = _.map(tokens, function (elw, ilw) { return elw == data['tokens'][idata+ilw] })
        return _.every(matchLabelWords, _.identity)
      })
      
      // concat 'begin' & 'in' data['labels'] right now
      var count = 0
      _.each(matchArray, function (match, idx){
        count = ((match == true)||(count>0 && count<tokens.length)) ? count+1 : 0
        if (count == 1)
          data['labels'][idx] = data['labels'][idx].concat("b_" + label)
        else if (count>0 && count<=tokens.length)
          data['labels'][idx] = data['labels'][idx].concat("i_" + label)
      })   
    })
  })

  // concat 'other' to data['labels']
  _.each(data['labels'], function (e, i) {
    if (e.length == 0)
      data['labels'][i] = data['labels'][i].concat("other")
  })

  return data
}

BIOLabel.prototype.getLabelFromSequence = function(tags, tokens) {
  var START = 'START'
  var STOP = 'STOP'
  var all_predicted_tags = _.flatten(tags)
  var all_tokens = _.flatten(tokens)
  var entities = {}
  tags.forEach(function (sequence, idx) {
    var state = STOP
    var entity = []
    var predicted_tag = ''
    sequence.forEach(function (tag, jdx) {
      if (tag.match(/other/) || predicted_tag != tag.slice(2)) {
        if (!_.isEmpty(entity)) {
          entities[predicted_tag] = (_.isUndefined(entities[predicted_tag])) ? [] : entities[predicted_tag]
          entities[predicted_tag].push(entity.join(' '))
        }
        entity = []
        state = STOP
      }
      if (tag.match(/b_.*/)) {
        predicted_tag = tag.slice(2)
        state = START
      }
      if (state == START)
        entity.push(tokens[idx][jdx])
    })
    if (entity.length > 0)
      entities[predicted_tag].push(entity.join(' '))
  })
  var valid_tags = Object.keys(entities)
  valid_tags.forEach(function (t) { entities[t] = _.uniq(entities[t]) })
  return entities
}

module.exports = new BIOLabel ()