Research On The Application Of Sentiment Computing In Japanese Anime Reviews

Text sentiment computing,a technique that focuses on the correspondence between textual information and sentiment states,has become an important preventive tool in monitoring and early warning tasks in the security field.Japanese anime has the subtle influence on the concept and consciousness of the audience in the process of dissemination.In combination with the background of the network era,the association between Japanese anime,audience views and communication influence is constructed.The emotional data in Japanese anime comments as the initial exploration of audience emotional experience for public opinion management of Japanese anime communication is conducive to promoting the construction of the network communication platform.However,with the widespread use of current online social platforms,user comments present the more colloquial and non-standardized form of expression.The traditional classification models are no longer suitable for the personalized expressions on current online platforms.To address the problems of overreliance on text sequences,neglecting of syntactic structure,and the poor interpretability of the feature space,a sentiment classification algorithm based on the Graph Neural Network is proposed.First,an external syntactic system is introduced to construct semantic graph of short-text,and the task of text classification is transformed into graph classification.Second,a graph filter based on the spatial-domain is constructed to capture the semantic and structural features of the graph.Third,a Long Short-Term Memory Network is introduced as the state updater to filter out the noisy information.Finally,through the operation of read-out,graph-level feature is read out and used for the classification task.This study provides the new research perspective for the current non-normalizedsentence-level classification task.It further extends the practical application scenarios of Graph Neural Networks in natural language processing tasks.At the same time,a sentiment analysis system based on the Graph Neural Network is constructed and applied to Japanese anime reviews to analyze tendencies of audience sentiment.The algorithm is used to correlate Japanese manga works,audiences’sentiments and communication influences in order to study the communication security of works by taking audiences’sentiments as feedback.The main innovations and contributions of this paper are as follows.（1）Aiming at the problem of traditional classification models,such as neglecting of syntactic structure,it researches and constructs a sentiment classification model based on the Graph Neural Network in order to improve the interpretability of feature space.（2）Through the construction of the graph filter based on spatial-domain,the information sharing among neighborhood nodes is realized.（3）The Long Short-Term Memory Network is introduced as the state updater to filter out the noise information while the node features are updated.（4）The lossed strategies of node features and edge features and the combined strategy of low-order adjacent features and high-order adjacent features are used to optimize the problems of over-smoothing and assumption of strong homogeneity.The experimental results show that the proposed model called GNN-LSTM（Graph Neural Networks and Long Short-Term Memory）achieves 95.25%accuracy and95.22%F1-score on the public dataset of microblog comments.Then,the optimized model called N₂GNN-LSTM is proposed.It obtains 95.58%accuracy and 95.51%F1-score.Meanwhile,the performance of the model is tested on other different comment datasets.It outperforms benchmark methods,proving that the sentiment analysis algorithm based on Graph Neural Networks has good ability to extend on online comment area of the Internet.