Sentiment Analysis Based On Deep Learning And Rhetorical Relation

In natural language process,sentiment Analysis is applied to detect whether subjective opinions are expressed in the natural language text and what exactly it is.A typical task of sentiment analysis is to classify the categories of sentiments,including positive,negative and neutral sentiment.Once the emotional tendencies are obtained,the applications in domain of business,market and consensus can be extended.According to the level of sentiment analysis,missions are supposed to be classified into lexicon level,phrase level,sentence level,document level,etc.The subject investigated in our paper is sentence level sentiment analysis in English language.Our paper proposes a deep learning and rhetorical relation based sentiment analysis method.On the basis of Rhetorical Structure Theory,our method splits the entire sentences into several segments in term of rhetorical relation,each segment has a unique role in hierarchy type(nucleus or satellite unit).Considering the semantic information,hierarchy type and rhetorical structure,our model finally aggregates the sentiment scores for entire sentences.A special attention mechanism related with rhetorical relation is designed in our model and exploited to extract the important semantic information from the text.Multiple models are implemented to independently handle sentiment analysis tasks in different granularity,enabled each model to focus on the certain aspect of semantic information.The accuracy and stability of our model in sentiment analysis is proven via experiments among variant deep learning model.