All files / lib/nlp nlp-classifier.js

92.98% Statements 106/114
84.81% Branches 67/79
100% Functions 18/18
92.98% Lines 106/114
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313                                              31x 31x 31x                                       186x 186x 5x   186x 84x   186x 84x   186x 186x   186x 185x   186x 185x   186x 185x   186x 186x                 2165x                             463x 463x 463x 5975x 5975x       16x     447x                   453x                 455x 2x   453x 2x   451x 451x 451x 8x   443x 443x 443x 2314x                   8x 1x   7x 7x 7x 1x   6x 6x 4x 4x 10x 10x 10x                             923x     923x 923x 923x 19022x   923x       373x 373x 2141x 2063x 2063x 2051x   12x       373x             47x 47x 47x 321x 321x         47x 39x     47x 47x 47x 321x 321x         47x         602x 117x 52x     550x       52x 52x 441x   52x 52x 52x 441x         52x                     602x 602x 602x     602x     602x 550x   52x           52x 51x 38x     14x                                         2x       31x  
/*
 * Copyright (c) AXA Shared Services Spain S.A.
 *
 * Permission is hereby granted, free of charge, to any person obtaining
 * a copy of this software and associated documentation files (the
 * "Software"), to deal in the Software without restriction, including
 * without limitation the rights to use, copy, modify, merge, publish,
 * distribute, sublicense, and/or sell copies of the Software, and to
 * permit persons to whom the Software is furnished to do so, subject to
 * the following conditions:
 *
 * The above copyright notice and this permission notice shall be
 * included in all copies or substantial portions of the Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE
 * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
 * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
 * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 */
 
const NlpUtil = require('./nlp-util');
const LogisticRegressionClassifier = require('../classifiers/logistic-regression-classifier');
const BinaryNeuralNetworkClassifier = require('../classifiers/binary-neural-network-classifier');
 
/**
 * Class for the NLP Classifier.
 * In the settings you can specify:
 * - classifier (optional): The Machine Learning Classifier Class. If not
 *      provided, then a default Logistic Regression Classifier is used.
 * - stemmer (optional): The language stemmer (also tokenize). If not
 *      provided, you can provide the language and the default stemmer
 *      for this language will be used.
 * - language (optional): If you don't provide a stemmer, then you can
 *      provide a language so a default stemmer for this language will
 *      be used.
 */
class NlpClassifier {
  /**
   * Constructor of the class.
   * @param {Object} settings Settings for this instance.
   */
  constructor(settings) {
    this.settings = settings || {};
    if (!this.settings.language) {
      this.settings.language = 'en';
    }
    if (this.settings.useNeural === undefined) {
      this.settings.useNeural = true;
    }
    if (this.settings.useLRC === undefined) {
      this.settings.useLRC = true;
    }
    Eif (!this.settings.classifier && this.settings.useLRC) {
      this.settings.classifier = new LogisticRegressionClassifier();
    }
    if (!this.settings.neuralClassifier && this.settings.useNeural) {
      this.settings.neuralClassifier = new BinaryNeuralNetworkClassifier();
    }
    if (!this.settings.stemmer) {
      this.settings.stemmer = NlpUtil.getStemmer(this.settings.language);
    }
    if (this.settings.keepStopWords === undefined) {
      this.settings.keepStopWords = true;
    }
    this.docs = [];
    this.features = {};
  }
 
  /**
   * Generate the vector of features.
   * @param {String} utterance Input utterance.
   * @returns {String[]} Vector of features.
   */
  tokenizeAndStem(utterance) {
    return typeof utterance === 'string'
      ? this.settings.stemmer.tokenizeAndStem(
          utterance,
          this.settings.keepStopWords
        )
      : utterance;
  }
 
  /**
   * Gets the position of a utterance for an intent.
   * @param {Object} srcUtterance Utterance to be found.
   * @param {Object} intent Intent of the utterance.
   * @returns {Number} Position of the utterance, -1 if not found.
   */
  posUtterance(srcUtterance, intent) {
    const utterance = this.tokenizeAndStem(srcUtterance);
    const utteranceStr = utterance.join(' ');
    for (let i = 0; i < this.docs.length; i += 1) {
      const doc = this.docs[i];
      if (
        doc.utterance.join(' ') === utteranceStr &&
        (!intent || doc.intent === intent)
      ) {
        return i;
      }
    }
    return -1;
  }
 
  /**
   * Indicates if an utterance already exists, at the given intent or globally.
   * @param {String} utterance Utterance to be checked.
   * @param {String} intent Intent to check, undefined to search globally.
   * @returns {boolean} True if the intent exists, false otherwise.
   */
  existsUtterance(utterance, intent) {
    return this.posUtterance(utterance, intent) !== -1;
  }
 
  /**
   * Adds a new utterance to an intent.
   * @param {String} srcUtterance Utterance to be added.
   * @param {String} srcIntent Intent for adding the utterance.
   */
  add(srcUtterance, srcIntent) {
    if (typeof srcUtterance !== 'string') {
      throw new Error('Utterance must be an string');
    }
    if (typeof srcIntent !== 'string') {
      throw new Error('Intent must be an string');
    }
    const intent = srcIntent.trim();
    const utterance = this.tokenizeAndStem(srcUtterance);
    if (utterance.length === 0 || this.existsUtterance(utterance)) {
      return;
    }
    const doc = { intent, utterance };
    this.docs.push(doc);
    utterance.forEach(token => {
      this.features[token] = (this.features[token] || 0) + 1;
    });
  }
 
  /**
   * Remove an utterance from the classifier.
   * @param {String} srcUtterance Utterance to be removed.
   * @param {String} srcIntent Intent of the utterance, undefined to search all
   */
  remove(srcUtterance, srcIntent) {
    if (typeof srcUtterance !== 'string') {
      throw new Error('Utterance must be an string');
    }
    const intent = srcIntent ? srcIntent.trim() : undefined;
    const utterance = this.tokenizeAndStem(srcUtterance);
    if (utterance.length === 0) {
      return;
    }
    const pos = this.posUtterance(utterance, intent);
    if (pos !== -1) {
      this.docs.splice(pos, 1);
      utterance.forEach(token => {
        this.features[token] = this.features[token] - 1;
        Eif (this.features[token] <= 0) {
          delete this.features[token];
        }
      });
    }
  }
 
  /**
   * Given an utterance, tokenize and steam the utterance and convert it
   * to a vector of binary values, where each position is a feature (a word
   * stemmed) and the value means if the utterance has this feature.
   * The input utterance can be an string or an array of tokens.
   * @param {String} srcUtterance Utterance to be converted to features vector.
   * @returns {Number[]} Features vector of the utterance.
   */
  textToFeatures(srcUtterance) {
    const utterance = Array.isArray(srcUtterance)
      ? srcUtterance
      : this.tokenizeAndStem(srcUtterance);
    const keys = Object.keys(this.features);
    const result = [];
    keys.forEach(key => {
      result.push(utterance.indexOf(key) > -1 ? 1 : 0);
    });
    return result;
  }
 
  tokensToNeural(tokens) {
    const result = {};
    for (let i = 0; i < tokens.length; i += 1) {
      if (this.features[tokens[i]]) {
        const value = Number.parseInt(tokens[i], 10);
        if (Number.isNaN(value)) {
          result[tokens[i]] = 1;
        } else {
          result['%number%'] = 1;
        }
      }
    }
    return result;
  }
 
  /**
   * Train the classifier with the existing utterances and intents.
   */
  async train() {
    Eif (this.settings.useLRC) {
      this.settings.classifier.clear();
      this.docs.forEach(doc => {
        const tokens = this.tokenizeAndStem(doc.utterance);
        this.settings.classifier.addObservation(
          this.textToFeatures(tokens),
          doc.intent
        );
      });
      if (this.settings.classifier.observationCount > 0) {
        await this.settings.classifier.train();
      }
    }
    Eif (this.settings.useNeural) {
      const corpus = [];
      this.docs.forEach(doc => {
        const tokens = this.tokenizeAndStem(doc.utterance);
        corpus.push({
          input: this.tokensToNeural(tokens),
          output: doc.intent,
        });
      });
      await this.settings.neuralClassifier.trainBatch(corpus);
    }
  }
 
  isEqualClassification(classifications) {
    for (let i = 0; i < classifications.length; i += 1) {
      if (classifications[i].value !== 0.5) {
        return false;
      }
    }
    return true;
  }
 
  normalizeNeural(classifications) {
    let total = 0;
    for (let i = 0; i < classifications.length; i += 1) {
      total += classifications[i].value;
    }
    Eif (total > 0) {
      const result = [];
      for (let i = 0; i < classifications.length; i += 1) {
        result.push({
          label: classifications[i].label,
          value: classifications[i].value / total,
        });
      }
      return result;
    }
    return classifications;
  }
 
  /**
   * Get all the labels and score for each label from this utterance.
   * @param {String} utterance Utterance to be classified.
   * @returns {Object[]} Sorted array of classifications, with label and score.
   */
  getClassifications(utterance) {
    const tokens = this.tokenizeAndStem(utterance);
    Eif (this.settings.useLRC) {
      const classificationLRC = this.settings.classifier.getClassifications(
        this.textToFeatures(tokens)
      );
      Iif (!this.settings.useNeural) {
        return classificationLRC;
      }
      if (this.isEqualClassification(classificationLRC)) {
        return classificationLRC;
      }
      const classificationNeural = this.normalizeNeural(
        this.settings.neuralClassifier.classify(
          this.tokensToNeural(tokens),
          true
        )
      );
      if (classificationLRC[0].label === classificationNeural[0].label) {
        if (classificationNeural[0].value < classificationLRC[0].value) {
          return classificationLRC;
        }
      }
      return classificationNeural;
    }
    if (this.settings.useNeural) {
      const classification = this.settings.neuralClassifier.classify(
        this.tokensToNeural(tokens),
        true
      );
      if (this.isEqualClassification(classification)) {
        return classification;
      }
      return this.normalizeNeural(classification);
    }
    return [];
  }
 
  /**
   * Given an utterance, get the label and score of the best classification.
   * @param {String} utterance Utterance to be classified.
   * @returns {Object} Best classification of the observation.
   */
  getBestClassification(utterance) {
    return this.getClassifications(utterance)[0];
  }
}
 
module.exports = NlpClassifier;