结合主题和分层注意混合网络的文本情感分析  

Text Sentiment Analysis of Combining Topic and Hierarchical Attention Hybrid Networks

作　　者：饶冬章 任淑霞[1] 赵宗现 

机构地区：[1]天津工业大学计算机科学与技术学院,天津

出　　处：《计算机科学与应用》2022年第11期2451-2459,共9页Computer Science and Application

摘　　要：文本情感分析一直以来都是自然语言处理的研究热点,近几年,深度神经网络在文本情感分析任务中取得了不错的效果。尽管取得了进展,但提出的模型没有利用整个语料库的统计信息,也没有将文档的体系结构的知识纳入到模型中。针对上述问题,本文提出了一种结合主题和分层注意混合网络的文本情感分析模型。首先,利用主题模型对数据集的主题进行提取,并结合文本的词嵌入和句子嵌入来丰富特征空间,以解决传统神经网络无法融入数据统计信息的问题;然后,采用卷积神经网络来降低特征空间的维度,同时,学习关键的主题信息;最后,使用带有主题感知的分层注意网络对模型进行训练,来关注文本中更重要的单词和句子。实验结果表明,提出的模型具有更好的分类性能,能够更好地揭示文本的情感。Text sentiment analysis has long been a hot research topic in natural language processing, and in recent years, deep neural networks have achieved good results in text sentiment analysis tasks. Despite the progress made, the proposed models do not make use of statistical information from the whole corpus, nor do they incorporate knowledge of the architecture of the documents into the models. To address these issues, this paper proposed a model for text sentiment analysis that combines thematic and hierarchical attention hybrid networks. Firstly, a topic model is used to extract the topics of the dataset and combine the word embeddings and sentence embeddings of the text to enrich the feature space to solve the problem that traditional neural networks cannot incorporate statistical information from the data;then, a convolutional neural network is used to reduce the dimensionality of the feature space while learning key topic information;finally, the model is trained using a hierarchical attention network with topic awareness to focus on more important words and sentences in the text. The experimental results show that the proposed model has better classification performance and can better reveal the sentiment of the text.

关 键 词：文本情感分析 主题模型 分层注意力机制 混合神经网络 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]