Research on Sentiment Analysis of Tibetan Short Text Based on Dual-channel Hybrid Neural Network

In response to the problem of varying degrees of loss of textual semantic information with the increase of model depth in a single-channel hybrid neural network model, this paper proposes a dual-channel hybrid neural network model-ALDCBAT based on the idea of multi-channel hybrid neural network, usi...

Full description

Saved in:
Bibliographic Details
Published in2023 IEEE 4th International Conference on Pattern Recognition and Machine Learning (PRML) pp. 377 - 384
Main Authors Zhu, Yulei, Luosai, Baima, Zhou, Liyuan, Qun, Nuo, Nyima, Tashi
Format Conference Proceeding
LanguageEnglish
Published IEEE 04.08.2023
Subjects
Analytical models
BiGRU
Feature extraction
Logic gates
Machine learning
Neural networks
pretraining model
Semantics
Sentiment analysis
TextCNN
Tibetan sentiment analysis
Online AccessGet full text

Cover

More Information
Summary:In response to the problem of varying degrees of loss of textual semantic information with the increase of model depth in a single-channel hybrid neural network model, this paper proposes a dual-channel hybrid neural network model-ALDCBAT based on the idea of multi-channel hybrid neural network, using ALBERT pre-training model, convolutional neural network and bidirectional gated unit network. The model first vectorizes Tibetan texts using ALBERT pre-training model, and then inputs the word vectors into the TextCNN model and the BiGRU model respectively. Secondly, an attention mechanism is introduced to enhance the text feature extraction ability of the BiGRU model. Finally, the output of the TextCNN model is concatenated with the output of the BiGRU model and the attention mechanism as the final output. Experimental results show that the classification accuracy of the dual-channel hybrid neural network model proposed in this paper is 91.12%, which partly solves the problem of loss of semantic information and effectively improves the classification accuracy of Tibetan sentiment analysis.
DOI:10.1109/PRML59573.2023.10348366