A graph attention network under probabilistic linguistic environment based on Bi-LSTM applied to film classification

Film reviews contain rich and complex linguistic information that can reflect the opinions and emotions of the reviewers. However, existing methods for emotion classification of film reviews rely on quantifying qualitative evaluations numerically. This approach can lead to difficulties in interpretation, information loss, and performance degradation under massive data. In this paper, we propose a novel method that utilizes a probabilistic linguistic term set (PLTS) and graph attention network (GAT) to classify films based on their emotional content in long reviews. Firstly, the Bi-directional long short-term memory (Bi-LSTM) method is used to convert film reviews into distributed emotional probabilities. This approach not only captures the emotional information in reviews, but also avoids the limitations of numerical quantification. Secondly, using PLTS to represent emotional information not only considers the relationships of linguistic features but also captures multiple emotional information simultaneously. Finally, we utilize multiple GATs to learn and aggregate the distributed emotional probabilities, enabling our method to fully perceive multiple emotional information in the reviews. Experimental results demonstrate that our method outperforms other models in classification accuracy on the IMDB film review dataset. Our method emulates human thinking to analyze emotional information in reviews and uses a human-like attention mechanism to learn the interrelationship between emotions in film reviews. Therefore, our method exhibits significant improvements in both accuracy and interpretability compared to current models, making it applicable to diverse domains that necessitate the analysis of linguistic data. Overall, the proposed method in this paper presents a novel and effective approach to analyzing and classifying films based on linguistic reviews.