Text Sentiment Classification Based On Sentence Structure Information

With the development of natural language processing technology and deep learning,various automatic scoring systems have emerged endlessly.In the automatic scoring of subjective questions,especially in distinguishing and analyzing questions,the answerer needs to analyze and answer characters or events from various angles and levels,and analyze their impact and evaluation.This approach is very similar to goal-based sentiment analysis.Target-based sentiment analysis is a sub-task of natural language processing.It classifies the sentiment tendency of target vocabulary in specific sentences,and evaluates it as positive,neutral or negative.Therefore,in the automatic scoring of subjective questions,the person who scores the answer to the analysis question can be transformed into a judgment of the emotion classification score based on various aspects of the answer to the analysis question.This paper proposes a neural network model that uses sentence structure to assist information to classify sentiment levels at the aspect level.In the existing aspect-level sentiment classification tasks,the traditional classification method based on sentiment dictionary and machine learning cannot meet people's requirements for the accuracy of sentiment classification results.In deep learning,the commonly used method is to send the sentence to the encoder for encoding and representation as a semantic information vector,and then use related technologies(such as the attention mechanism)to extract the feature relationship between the semantic vector and the target vocabulary(aspect).The classifier classifies the extracted relational features.However,these commonly used methods often deal with the semantic information of the text,thus ignoring the information contained in the text sentence structure(such as the modifiers of the feature words often carry emotional information).Therefore,this paper proposes a neural network model that uses both text semantic information and sentence structure information.This paper uses Transformer network to model text semantic information and uses Bi-LSTM network to extract sentence structure information.Then,by using The attention mechanism constructs the text semantic information and sentence structure information into a text representation based on the sentence structure information,and finally sends it to the classifier for classification.At the end of the article,experiments on the Sem Eval 2014 dataset prove that networks with text sentence structure information have higher accuracy than networks that use only semantic information.