activity-stream/lib/NmfTextTagger.jsm

63 строки
2.0 KiB
JavaScript

/* This Source Code Form is subject to the terms of the Mozilla Public
* License, v. 2.0. If a copy of the MPL was not distributed with this
* file, You can obtain one at http://mozilla.org/MPL/2.0/. */
"use strict";
const {toksToTfIdfVector} = ChromeUtils.import("resource://activity-stream/lib/Tokenize.jsm", {});
this.NmfTextTagger = class NmfTextTagger {
constructor(model) {
this.model = model;
}
/**
* A multiclass classifier that scores tokenized text against several classes through
* inference of a nonnegative matrix factorization of TF-IDF vectors and
* class labels. Returns a map of class labels as string keys to scores.
* (Higher is more confident.) All classes get scored, so it is up to
* consumer of this data determine what classes are most valuable.
*/
tagTokens(tokens) {
let fv = toksToTfIdfVector(tokens, this.model.vocab_idfs);
let fve = Object.values(fv);
// normalize by the sum of the vector
let sum = 0.0;
for (let pair of fve) {
// eslint-disable-next-line prefer-destructuring
sum += pair[1];
}
for (let i = 0; i < fve.length; i++) {
// eslint-disable-next-line prefer-destructuring
fve[i][1] /= sum;
}
// dot the document with each topic vector so that we can transform it into
// the latent space
let toksInLatentSpace = [];
for (let topicVect of this.model.topic_word) {
let fvDotTwv = 0;
// dot fv with each topic word vector
for (let pair of fve) {
let [termId, tfidf] = pair;
fvDotTwv += tfidf * topicVect[termId];
}
toksInLatentSpace.push(fvDotTwv);
}
// now project toksInLatentSpace back into class space
let predictions = {};
Object.keys(this.model.document_topic).forEach(topic => {
let score = 0;
for (let i = 0; i < toksInLatentSpace.length; i++) {
score += toksInLatentSpace[i] * this.model.document_topic[topic][i];
}
predictions[topic] = score;
});
return predictions;
}
};
const EXPORTED_SYMBOLS = ["NmfTextTagger"];