Emotional Classification Of Film Reviews Based On ERNIE Model

As Internet technology develops,modern film industry has formed the mode of online ticket purchase,offline viewing and online film review and feedbackFilm reviews,which are complicated and variable,reflect audiences’ attitudes.The deep semantic mining and emotional classification of online film reviews can help merchants or service providers make relevant decisions,which has important commercial value and academic significance.In recent years,with the rapid development of deep learning,pretraining neural network model based on large-scale corpus data has become the preferred model to solve natural language processing tasks.Among them,ERNIE model directly uses prior semantic knowledge,enhances the capacity of the semantic representation.Thus,ERNIE model has a great advantage in processing Chinese text task.In this thesis,the online review data of Chinese films obtained from Douban.com based on Python crawler technology is taken as the research object,and the emotional tendency of the text data is marked by combining the score from douban.com,so as to carry out targeted cleaning work on the film review data set.In order to improve the accuracy of Chinese word segmentation,the thesis collects film professional vocabulary and the new words in the film field as the supplement of jieba segmentation tool.We also visually analyze the feature of the data set which has been cleaned and do the word frequency stastistics on the comment text,so that the focus of online movie reviews can be displayed more directly.To solve the problem that semantic information can not be fully utilized in word vector representation in Chinese text classification method of film review,ERNIE pretraining model based on knowledge enhanced semantic representation is used to obtain distributed text vector representation,which is input into fully connected neural network to train emotional multi-classification of Chinese text.In the same experimental environment,the feasibility of the model is verified by comparison with FASTTEXT,BERT and other text classification algorithms based on Word2Vec word embedding and BERT word embedding.According to the experimental results,compared with BERT model,the classification accuracy,macromean recall rate and MacroFl value of ERNIE model are improved by about 1%,1.23%and 1.01%respectively.It proves that in the classification problem of medium and long Chinese text,ERNIE model has stronger semantic feature extraction ability than Word2Vec,BERT and other word embedding methods,which provides a reference for subsequent research in the field of natural language processing.