Word vector training method and device

Abstract

The invention provides a word vector training method and a word vector training device. The word vector training method includes: acquiring a newly occurring word bank, wherein words in the newly occurring word bank and words in an old word bank form a new word bank, and words in the old word bank correspond to old word vectors; performing initialization treatment on the words in the newly occurring word bank so as to achieve the purposes that word vectors of the words in the newly occurring word bank, which belong to the old word bank, are the old word vectors, and word vectors of the words in the newly occurring word bank, which only belong to the newly occurring word bank, are random word vectors; respectively updating the word vectors of the words in the newly occurring word bank according to noise distribution corresponding to the old word bank and noise distribution corresponding to the newly occurring word bank. The word vector training method and the word vector training device reduce calculation amount in training of the word vectors.

Description

Technical field [0001] The present invention relates to machine learning technology, in particular to a word vector training method and device. Background technique [0002] In machine learning technology, in order to make the machine understand the meaning of human language, the word representation tool of the neural network language model converts each vocabulary in the human language into the form of word vector, so that the computer can learn the human language through the word vector The meaning of each word. [0003] In the prior art, the word representation tool obtains the word vector of each vocabulary by learning all the words in the vocabulary. [0004] With the existing technology, when a new vocabulary is added to the vocabulary, because the frequency of each word has changed, it is necessary to relearn all the vocabulary in the new vocabulary to obtain new word vectors for each vocabulary, and to train the words The amount of calculation is large when the vector is us...

AI Technical Summary

Problems solved by technology

[0004] Using the existing technology, when a new vocabulary is added to the vocabulary, because the frequency of each word has changed, it is necessary to re-learn all the vocabulary in the new vocabulary to obtain new word vectors for each vocabulary, training words The amount of calculation is large when vector

Images