Label representation method, terminal device and storage medium based on metric emotion learning

Abstract

The invention discloses a label representation method, terminal device and storage medium based on metric emotion learning. The method includes: acquiring a label set including a plurality of triplets, the triplets including a first vocabulary, an antonym of the first vocabulary and a second vocabulary A synonym for a vocabulary; determine the transformation matrix, use the transformation matrix to convert the triplet in the label set into a triplet in the vector space; determine each label in the text to be represented according to the likelihood ratio function and the triplet in the vector space tag words. The present invention corrects the relative positions between word vectors by using the conversion matrix, solves the overlapping problem of word vectors in the vector space, and makes the generated label words significantly improve the performance of the model in downstream tasks; at the same time, the present invention also uses The emotion transfer operation is used to obtain label words that can effectively represent the intermediate category, so that the accuracy of the final label words is close to the accuracy of manually setting label words, and the accuracy is greatly improved.

Description

technical field [0001] The invention belongs to a tag representation method, in particular to a tag representation method based on metric emotion learning, a terminal device and a computer-readable storage medium. Background technique [0002] Label Representation essentially replaces category labels with specific words to improve the model's understanding of text semantics. This method is widely used in many state-of-the-art transfer learning models to convert other natural language processing tasks into inference tasks, and has brought significant improvements to many tasks with fewer training samples. The generation of label words depends on the input data of a specific task. For example, the label set contained in the Yelp-Review dataset is {1, 2, 3, 4, 5}, and it is possible to artificially match a word for each label, that is, {" Great", "Good", "Okey", "Bad", "Terrible"}, and according to such settings, the classification model is converted from predicting numerical ...

AI Technical Summary

Problems solved by technology

[0005] The present invention provides a label representation method, terminal equipment and computer-readable storage medium based on metric emotion learning, thereby solving the technical problem of inaccurate automatically generated label words in the prior art

Images