Research on Sentiment Classification Method Base on Multi-objective Evaluation Subject

Research on Sentiment Classification Method Base on Multi-objective Evaluation Subject

In public opinion events, netizens often comment on multiple subjects involved in public opinion events from multiple perspectives. However, in traditional research methods, due to the lack of fine-grained classification of these evaluation subjects, the sentiment classification of netizens’ evaluation in public opinion events is not precise and accurate. This paper proposes an emotion classification method based on multi-objective evaluation subjects. Firstly, the method combines dependency parsing to identify the emotional words and different evaluation subjects in the comment text; Secondly, use the semantic relationship and emotional rules between the comment text to segment and associate the various emotional tendencies of different evaluation subjects in the sentence; Finally, long-short term memory neural networks are used to classify the emotions of different evaluation subjects in the same event. Using four types of review text as a data set, the experimental results of books show that compared with the traditional LSTM model method, the precision rate, recall rate and F1 value of the MO-LSTM sentiment classification method are improved by 6.6%, 7.9% and 5.5% respectively. This method can accurately identify the emotional tendency and help find the root causes of negative emotional tendencies of public opinion in the event analysis.