一种融合上下文特征的中文隐式情感分类模型

A Chinese implicit sentiment classification model combining contextual features

对网络上海量的文本数据进行情感分析,可以更好地挖掘网民行为规律、帮助决策机构了解舆情倾向和改善商家服务质量。在实际表达中,人们除了采用带有明显情感词的主观表达外,还采用含蓄的方式表达自己的主观倾向。带有显式情感词的文本情感分析作为自然语言处理领域的基础性研究任务,已经取得了丰富的研究成果。然而,针对隐式文本的情感分析技术还处于起步阶段。与显式情感分析任务相比,隐式情感分类任务更加困难。隐式表达文本具有中立性表达、缺乏情感词和上下文依赖的特点,使得传统的文本分类方法不再适用。针对以上问题,采用word2vec词嵌入技术提取文本特征,分别进行了基于TextCNN、LSTM和BiGRU分类模型的研究。在各个深度分类模型研究基础上,还进行了融合注意力机制的分类模型研究。针对隐式表达对上下文内容依赖的特点,设计了一种融合上下文语义特征和注意力机制的分类模型,增强了部分中立性隐式表达句的分类效果。最后在SMP2019公开数据集上进行了实验,取得了比上述几种基础深度网络模型与融合注意力机制分类模型更好的分类效果。

Sentiment analysis of the massive amount of text data on the network can better explore the behavior of netizens,help decision-making institutions understand the tendency of public opinion and improve the quality of service.In daily expression,in addition to adopting subjective expressions with obvious sentimental words,people also express their subjective tendencies in an implicit way.As a basic research task in the field of natural language processing,text sentiment analysis with explicit sentimental words has achieved rich research results.However,sentiment analysis techniques for implicit text are still in their infancy.Implicit sentiment classification tasks are more difficult than explicit sentiment analysis tasks.Implicitly expressed text has the characteristics of neutral expression,lacking of sentimental words,and context dependence,making the traditional text classification method no longer applicable.To cure the above problems,word2vec word embedding technology is used to extract text features,and the research based on TextCNN,LSTM and BiGRU classification models is carried out.Based on the research of various depth classification models,a classification model of the fusion attention mechanism is also studied.Aiming at the characteristics of implicit expression on contextual content,a classification model that combines contextual semantic features and attention mechanism is designed to enhance the classification effect of some neutral implicit expressions.Finally,experiments are carried out on the SMP2019 public dataset,and our proposal obtains better classification results than the aforementioned basic deep network model and the fusion attention mechanism classification model.