Research On The Method Of Fine-grained Sentiment Analysis Of Movie Reviews Based On Evaluation Objects

In recent years,as society has continued to evolve,people’s hunger for a better life has been increasing.Especially due to the COVID-19 pandemic in the past three years,which has increased people’s concern about choosing to watch movies,but also raised the audience’s requirements and standards for films.This requires us to make accurate sentiment analysis of the evaluation texts after watching a movie,so as to better help audiences choose more valuable films,help cinemas attract more audiences in the postpandemic era,and promote China’s cultural soft power.But nowadays,there is a huge amount of review text data and more and more valuable information,which requires a higher standard for text sentiment analysis.Since the traditional sentiment analysis techniques based on dictionaries and machine learning cannot handle the explosive amount of information in review texts.Therefore,this paper uses deep learning related techniques to conduct an in-depth study on movie reviews,which mainly includes the following two parts:(1)Study of CNN-BiLSTM-CRF based approach for evaluating object extraction.For the first subtask of fine-grained sentiment analysis,the proposed method utilizes a CNN-BiLSTM-CRF model to extract evaluation objects.The model utilizes CNN and BiLSTM two-layer parallelism to fully acquire the feature vector,CNN can efficiently acquire local patterns and features in text sequences,and BiLSTM can notice the contextual information.The parallel approach enhances the effectiveness of extracting evaluation objects.Finally,the model combines the parallel feature information and obtains the optimal result through discriminative sequence modeling.(2)Study of EO-BiLSTM-Att based method for analyzing the affective tendencies of evaluation subjects.In the second subtask of the fine-grained sentiment analysis based on the sentiment analysis task of the evaluation object’s text,the EO-BiLSTMAtt model is used in this paper.The evaluation object is embedded into the model and fused with the feature vectors in BiLSTM,which is more relevant and accurate for the results of sentiment tendency analysis.Subsequently,an attention mechanism is trained based on the fused vector to reduce the computational complexity caused by direct concatenation of vectors.This approach captures more important words by assigning feature weights,thereby sentiment analysis’ effectiveness is increased.According to experimental findings,the model’s accuracy has increased when compared to other models.