摘要
目前的神经网络一般只将词粒度层面的词向量作为输入,忽略了语义层面的全局语义特征.针对此问题,提出了一种基于局部特征和全局特征融合的情感分类方法,以解决评论特征稀疏和主题聚焦性差的问题.对于局部特征,选择基于情感词典和BiLSTM神经网络模型提取基于词向量的文本特征.对于文本集的全局主题特征,采用神经主题模型提取文本主题特征,并将其作为全局特征来表示短文本信息.最终将基于局部加权词向量的文本特征和基于神经主题模型的文本主题特征进行拼接,并通过Softmax层输出,完成文本情感分类.结果表明:融合全局主题语义和局部加权词向量可以更加丰富神经网络的特征,从而有效地提高情感分类的准确率.
At present,neural networks generally only take the word vector at the word granularity level as the input,but the global semantic features at the semantic level are ignored.Aiming at the above problem,an sentiment classification method based on the combination of local features and global features is proposed to solve the problem of sparse comment features and poor topic focus.For the local features of documents,sentiment lexicon and BiLSTM neural network model are used to extract text features features based on word vector.For the global topic features of document set,neural topic model is adopted to take the text topic feature as the global feature to represent the short text information.Finally,the text features based on locally weighted word vector and the text topic features based on neural topic model are spliced and the sentiment classification is completed through the output of Softmax layer.The results show that the combination of global topic semantic features and local word vector features can enrich the features of neural network,so as to effectively improve the accuracy of sentiment classification.
作者
胥桂仙
陈思瑾
孟月婷
张廷
于绍娜
XU Guixian;CHEN Sijin;MENG Yueting;ZHANG Ting;YU Shaona(School of Information Engineering,Minzu University of China,Beijing 100081,China)
出处
《中南民族大学学报（自然科学版）》
CAS
北大核心
2023年第4期526-534,共9页
Journal of South-Central University for Nationalities：Natural Science Edition
基金
国家社会科学基金资助项目(19BGL241)
教育部人文社会科学研究规划基金资助项目(18YJA740059)
中央民族大学交叉课题资助项目(2020MDJC05)。
关键词
情感分析
特征融合
神经主题模型
词向量
sentiment analysis
feature fusion
neural topic model
word vector
