Authorship Detection on Classical Chinese Text Using Deep Learning

Authorship detection has played an important role in social information science. In this study, we propose a support vector machine (SVM)-based authorship detection model for classical Chinese texts. Term frequency-inverse document frequency (TF-IDF) feature extraction technique is combined with the SVM-based method. The linguistic features used in this model are based on TF-DIF calculations of different function words, including literary Chinese words, end-function words, vernacular function words, and transitional function words. Furthermore, a bidirectional long short-term memory (BiLSTM)-based authorship model is introduced to detect authorship in classical Chinese texts. The BiLSTM model incorporates an attention mechanism to better capture the meaning and weight of the words. We conduct a comparative analysis between the SVM-based and BiLSTM-based models in the context of authorship detection in Chinese classical literature. The applicability of the two authorship detection models for classical Chinese texts is examined. Results indicate varying authorship between different sections of the texts, with the SVM model outperforming the BiLSTM model. Notably, these classification outcomes are consistent with findings from prior studies in classical Chinese literary analysis. The proposed SVM-based authorship detection model is especially suited for automatic literary analysis, which underscores its potential for broader literary studies.