Loss function optimization method, device and sample classification method for classification model

Abstract

Embodiments of the present application provide a method and device for optimizing a loss function of a classification model, and a sample classification method. Wherein, the optimization method includes: generating a filter vector corresponding to the classification label vector, the classification label vector and the filter vector both include the dimension corresponding to the first classification and the dimension corresponding to the second classification, and the second classification in the filter vector The corresponding dimension value is zero; the original loss function is generated according to the classification label vector and the output of the classification model; the original loss function is filtered using the filter vector to remove the second-class classification components in the original loss function, and the loss filter is obtained function; according to the preset rules, the loss filtering function is post-processed to obtain the loss optimization function. Therefore, the optimized loss function can increase the learning weight of the classification model for the text features of the first category, and does not learn the text features of the second category, reducing the generalization of the classification model and improving the accuracy of text classification.

Description

technical field [0001] The present application relates to the technical field of natural language processing, in particular to a method and device for optimizing a loss function of a classification model, and a sample classification method. Background technique [0002] In the field of natural language processing technology, the TextCNN (TextConvolutional Neural Network, text convolutional neural network) model is a mainstream solution for text classification models. The principle of the TextCNN model is to convolve text features through a convolutional neural network. During the convolution process, multiple convolution kernels are used to extract important text features from the text, and text classification is performed based on the extracted text features. [0003] In the prior art, the TextCNN model has no difference in the feature extraction of the input text, that is to say, every time each text enters the TextCNN model, the TextCNN model will learn the text once and ...

AI Technical Summary

Problems solved by technology

[0004] The embodiment of the present application provides a classification model loss function optimization method, device and sample classification method to solve the problem of the generalization ability of the classification model in the prior art due to the indiscriminate learning of classification features.

Images