This study is based on the IMDB film review dataset released by Stanford University, and constructs a film review analysis model that integrates traditional machine learning and deep learning, with statistical methods as the main approach, to achieve sentiment classification and theme modeling of English film reviews. In the emotion classification section, this article focuses on using statistical feature extraction and modeling methods. Firstly, spaCy is used to perform preprocessing such as word segmentation and form restoration. Then, the TF-IDF algorithm is used to extract text features, and a statistical classification model based on logistic regression is constructed. After grid search optimization, the model achieves an accuracy of 88.0% on the test set. Meanwhile, by combining Word2Vec word vectors with a bidirectional LSTM neural network to design a deep learning path, with the support of regularization mechanisms such as dropout, the accuracy was improved to 88.24%. Finally, based on the validation set, a fusion strategy was proposed to combine the logistic regression and LSTM model prediction results with a weighted fusion of 0.45: 0.55, further improving the accuracy to 90.2%. In terms of topic modeling, the study adopts the classic LDA model to divide comment texts into 15 topic categories (such as post viewing sentiment, comedy, social issues, etc.), and analyzes the topic distribution of each comment through D-T matrix to achieve content dimension understanding beyond emotions.
Xu et al. (Thu,) studied this question.