LDA-Transformer Model in Chinese Poetry Authorship Attribution

Abstract

Authorship attribution broadly is defined as an analysis of individuals’ writing styles, which has been attracting a lot of interest. Although the problem has been widely explored, no previous studies attempt to identify Chinese classical poetry. In this paper, we presented a public classical poetry corpus in Tang Dynasty for Chinese authorship attribution. As a particular literal form, the theme feature plays a crucial role in Chinese poetry authorship attribution. To integrate the topic feature of the Chinese poem, we employed the latent Dirichlet allocation model to capture the extra theme information. Meanwhile, due to the incoherent expression of poetry text, it is hard to capture incoherence information effectively from Chinese poems. To tackle this problem, we propose a combination model called LDA-Transformer to perform authorship attribution of Chinese poetry. We conduct systematical evaluations for the proposed method on three Chinese poetry datasets. The experimental results suggest that the topic feature can effectively improve the performance of authorship attribution in Chinese poetry. Our model achieves state-of-the-art results on related baseline methods.