CINO双通道结合多头注意力机制藏文情感分类方法

Tibetan emotion classification based on CINO dual channel combined with multiple attention mechanism

为了解决藏文情感分类任务中现有的模型对文本语义信息理解和深层文本特征提取能力不足的问题,该文使用CINO(Chinese Minority PLM)预训练模型来获取动态词向量,通过TextCNN和BiGRU融合的双通道情感分类模型,分别实现获取文本局部特征和深层全局特征,并引入多头自注意力机制引导模型学习更重要的信息。实验结果表明,该文提出的双通道模型准确率高达92.84%,相较于该文的其他对比模型效果更佳。

In order to solve the problem of insufficient ability of existing models to understand text semantic information and extract deep text features in Tibetan sentiment classification task.In this paper,CINO(Chinese Minority PLM)pre⁃training model is used to obtain dynamic word vectors.Through the dual⁃channel sentiment classification model fused with TextCNN and BiGRU,local features and deep global features of the text are obtained respectively,and a multiple self⁃attention mechanism is introduced to guide the model to learn more robust information.The experimental results show that the accuracy of the dual channels model proposed in this paper is as high as 92.84%,which is better than other comparison models in this paper.