Aspect-Level Sentiment Classification Method Based on BERT and Multilayer Attention Mechanism

Abstract

The invention discloses an aspect-level emotion classification method based on BERT and a multi-layer attention mechanism, comprising the following steps: S1, preprocessing the training corpus; S2, establishing an aspect-level emotion classification model based on multi-layer attention; S3, Use the trained classification model to perform sentiment classification on the data to be predicted. The invention uses a multi-layer Attention mechanism to fully tap the correlation between aspcet words and DataContext words, and can effectively improve the accuracy of aspect-level emotion classification.

Description

technical field [0001] The present invention relates to an aspect-level sentiment classification method based on BERT and multi-layer attention mechanism. Background technique [0002] With the rise and development of social networks and e-commerce platforms, people prefer to post their comments online, such as comments on products on JD.com and Taobao, and comments on food on Meituan and Hungry. Merchants have valuable value. A comment may contain evaluations of multiple aspects. How to quickly obtain the sentiment polarity of aspect words has become an important research direction in the field of natural language processing. [0003] At present, there are mainly two processing methods: [0004] 1. Methods based on traditional machine learning, such as support vector machine (SVM) and logistic regression, extract features that depend on corpus implementation, and need to manually design features for data, and the quality of the model depends on the quality of feature desi...

AI Technical Summary

Problems solved by technology

[0007] (1) The method based on traditional machine learning needs to manually design complex features, which is very time-consuming

[0008] (2) Most of the existing deep learning-based methods use RNN to extract semantic information, such as: LSTN, CNN, although they have achieved good results, but compared with BERT to extract semantic information, there is still a certain gap

[0009] (3) The current model using the Attention mechanism is to calculate the attention of the Aspect as a whole to the Context word, without considering the attention of a single word in the Aspect to the Context word, which may easily lead to information loss

Images