A Cross-Domain Text Sentiment Classification Method Based on Domain Adversarial Adaptation

Abstract

The invention discloses a cross-domain text sentiment classification method based on domain confrontation self-adaptation, the method includes: inputting word vector matrix, category label and domain label of samples in source domain and target domain; using feature extraction based on convolutional neural network module, which extracts the low-level features of the sample; in the main task module, a constraint based on the distribution consistency of the source domain and the target domain is constructed, and the low-level samples are mapped to the regenerated kernel Hilbert space to learn high-level features with transferability; the source domain The high-level features are input into the category classifier, and on the basis of reducing the difference in the field, the classifier is guaranteed to have category discrimination for the samples; in the auxiliary task module, the domain invariance constraint based on adversarial learning is constructed, and the low-level features are input into the field with adversarial properties The classifier makes the classifier unable to distinguish the domain of the sample as much as possible, thereby extracting high-level features with domain invariance, which effectively solves the migration problem of the source domain classifier to the target domain.

Description

technical field [0001] The invention belongs to the technical field of text analysis, and in particular relates to a cross-domain text sentiment classification method based on domain confrontation self-adaptation. Background technique [0002] In recent years, with the vigorous development of artificial intelligence and machine learning technology, text sentiment classification technology has emerged as the times require. This technology can automatically classify the sentiment trend of text data, effectively solving the time-consuming and laborious problem of manual judgment. Traditional text sentiment classification methods usually use calibration data to train specific sentiment classifiers for a certain field to complete sentiment classification tasks. However, with the development of social media, the increasing number of new corpus gradually expands the scope of domains, and the amount of data in each domain is very large. Traditional text sentiment classification meth...

AI Technical Summary

Problems solved by technology

However, with the development of social media, the increasing number of new corpus gradually expands the scope of domains, and the amount of data in each domain is very large. Traditional text sentiment classification methods need to manually label a large amount of data for each newly added domain. Completing the training of the sentiment classifier, the process of manually labeling samples is inefficient

At the same time, with the passage of time and the development of society, the new feature words in the known field will gradually increase. Because there are certain differences in the feature distribution between the original sample and the new sample, the original sentiment classifier in this field will not be able to accurately predict the new comment data. emotional tendency

Images