融合音节和词条特征的藏文文本情感分类研究  

Tibetan Text Sentiment Classification Combining Syllables and Words

作　　者：孟祥和[1] 于洪志[1] MENG Xianghe;YU Hongzhi(Key Laboratory of China's Ethnic Languages and Information Technology of Ministry ofEducation,Northwest Minzu University,Lanzhou,Gansu 730030,China)

机构地区：[1]西北民族大学、中国民族语言文字信息技术教育部重点实验室,甘肃兰州730030

出　　处：《中文信息学报》2023年第2期80-86,共7页Journal of Chinese Information Processing

基　　金：西北民族大学2021年度中央高校基本科研业务费项目(31920210087);西北民族大学2020年度中央高校基本科研业务费项目(31920200116)。

摘　　要：将深度神经网络模型应用于藏文文本情感分类中,虽然取得不错的分类效果,但仍然存在因藏文评论文本长度较短引起的特征稀疏的问题,使得深度学习模型不能够提取到更为全面的藏文文本语义特征。该文提出一种以藏文音节和藏文词条同时作为文本基本表示对象,采用CNN、BiLSTM和Multi-Headed Self-Attention机制等深度学习模型完成对藏文评论文本情感分类的研究方法。实验首先对音节和词条进行向量化表示,然后分别采用多核卷积神经网络、BiLSTM和Multi-Headed Self-Attention机制获取藏文文本中多维度的内部特征,最后通过特征拼接,再经激活函数为Softmax的全连接神经网络完成文本情感分类。研究结果表明,在该文的实验测试语料集上,融合音节和词条特征模型的分类准确率要优于基于音节的模型和基于词条的模型。Deep neural network model has achieved good results in the sentiment classification of Tibetan texts.To further deal with the short Tibetan comment texts with sparse features,this paper proposes a method using both syllables and word as the unit of Tibetan text representation,which is fed into CNN,BiLSTM and Multi-Head Self-Attention mechanism to complete sentiment classification of Tibetan comment texts.Specifically,On the basis of the vectorization of syllables and words,we use multi-kernels convolutional neural networks,BiLSTM with Multi-Headed Self-Attention mechanism to obtain multi-dimensional internal features of the context in Tibetan text.Finally we concatenate the features and complete the text sentiment classification through the fully connected layer with the Softmax’s function.The experimental results show that,on the experimental test corpus of this paper,the accuracy of the proposed method is better than the syllables-based model and the words-based model.

关 键 词：藏文文本 情感分类 藏文音节 深度神经网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]