blob: 639c92b6e4af9ef62d8160058909010b8c648256 (
plain)
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
|
/* 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";
// We load this into a worker using importScripts, and in tests using import.
// We use var to avoid name collision errors.
// eslint-disable-next-line no-var
var EXPORTED_SYMBOLS = ["NmfTextTagger"];
const NmfTextTagger = class NmfTextTagger {
constructor(model, toksToTfIdfVector) {
this.model = model;
this.toksToTfIdfVector = toksToTfIdfVector;
}
/**
* 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 = this.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;
}
};
|
