Word vector computing method and word vector computing device based on electroencephalogram signal

Abstract

The invention provides a word vector computing method and a word vector computing device based on an electroencephalogram signal. The word vector computing method based on the electroencephalogram signal comprises the following steps: S1, collecting a text corpus, processing corpora in the text corpus to obtain corpora with a word group as a unit and in a continuous phrase format; S2, presenting the corpora in the continuous phrase format to a marker, so as to be read by the marker, and acquiring an electroencephalogram signal of the marker when reading each word group; and S3, taking the acquired electroencephalogram signal corresponding to the word group as a prediction target, training a word vector, predicting an electroencephalogram signal of context of the current word group by taking the current word group as a characteristic, and constructing a word vector representation model based on the electroencephalogram signal. According to the word vector computing method and a word vector computing device which are provided by the invention, word vector computing accuracy can be improved by virtue of the scheme.

Description

technical field [0001] The invention belongs to the technical field of natural language processing, and in particular relates to a word vector calculation method and device based on electroencephalogram signals. Background technique [0002] In natural language processing tasks, word vectors are usually used as the representation of words in the original text, so that numerical machine learning algorithms can be applied to text data. The basic idea of ​​the word vector model is: through a large number of corpus training, each word in a certain language is mapped into a fixed-length vector. Generally speaking, this length is much smaller than the size of the language dictionary, usually tens to hundreds dimension. All these vectors constitute a word vector space, and each vector can be a point in this space. By introducing the measure of "distance" in this space, the syntactic and semantic similarity between corresponding words can be judged according to the distance of wor...

AI Technical Summary

Problems solved by technology

[0004] 1. Only the attributes of the grammatical level of words are considered, and the attributes of the semantic level of words are not considered, so the word vectors usually trained can only express the relatively shallow relationship between words;

[0005] 2. Lack of modeling of human language cognitive process, ignoring important cognitive neuroscience and psychological characteristics;

[0006] 3. Due to the complexity of the human language cognition mechanism, the word vector obtained by simply predicting the context cannot reflect the characteristics of different natural language processing tasks, and the universality is poor

Images