摘要
在藏文自然语言处理领域内,目前情感分类的研究只有单一的文本和图像模态,采用方法也是传统的机器学习分类算法。然而评论数据一般是多模态的。本文选用基于神经网络的藏文情感分类(FCNNMSCTT)、情感表情图像分类(CNNMEITSA)、融合(FUSIONMODEL)三种模型对多模态藏语情感分类数据进行情感分类。最终实验结果是,FCNNMSCTT准确率达到了56%,CNNMEITSA准确率达到了88.75%。Fusion model准确率达到了96.98%。
In the field of Tibetan natural language processing, the research of emotion classification has only a single text and image mode, and the method used is also the traditional machine learning classification algorithm. However, comment data are generally multimodal. In this paper, three models of FCNNMSCTT, CNNMEITSA and FUSIONMODEL are used to classify the multimodal Tibetan emotion classification data. The experimental results show that the accuracy of FCNNMSCTT, CNNMEITSA,and FUSIONMODEL are 56%, 88.75%, and 96.98%, respectively.
作者
拉桑吉
安见才让
Ra Sangji;Anjian Cairang(School of computer,Qinghai University for Minzu,Xining,Qinghai 810007,China;Qinghai Key Laboratory of Tibetan information processing and machine translation;State Key Laboratory of Tibetan intelligent information processing and Application)
出处
《计算机时代》
2022年第10期98-102,共5页
Computer Era
基金
青海省科技计划项目(2019-ZJ-7066)
国家自然科学基金项目(61862054)
省部共建藏语智能信息处理及应用国家重点实验室(2021-Z-001)。
关键词
藏文情感分类
表情图像情感分类
神经网络模型
多模态数据集
Tibetan emotion classification
emotion classification of expression image
neural network model
multimodal data set
