A Novel Hand-Crafted Convolutional Neural Networks for Classification of Japanese Literature Corpus

Literary text categorization is essential in organizing and processing large corpora for research and educational purposes. Japanese literature, being highly diverse in both linguistic and stylistic aspects, poses special challenges to automated categorization. Current systems are mostly based on traditional machine learning methods or rule-based approaches, which tend to be weak at good feature extraction and context comprehension and thus yield moderate precision with poor generalizability. The earlier traditional models like support vector machines (SVM) and naïve Bayes classifiers, have been utilized but are found to be lacking when it comes to understanding the deep syntactic and semantic intricacies of literature. To address such inadequacies, this study suggests a hand-crafted convolutional neural network (HC-CNN)-inspired model specially tailored for Japanese literary text classification. The suggested architecture facilitates the effective extraction of hierarchical features from text representations, which leads to improved classification robustness and accuracy. Comparative evaluation using conventional classifiers proves the better performance of the proposed method. The findings show that suggested HC-CNN-model beats other methods with excellent predictive power and yields high accuracy of 96.30%, precision of 96.25%, recall of 96%, specificity of 96.15%, and F1-score of 96%. The results confirm the capability of deep learning methods to promote literary text classification and contribute to the construction of smart tools for digital humanities.