Bert-TiAttention: A Two-Input Model for Mining Hidden Authorial Writing Style Features

Mining the hidden authorial writing style features in text content has always been the main focus of author identification technology. Author identification models aim to learn unique and discriminative features to accurately identify the author of a text. In this paper, we propose a two-input single-output model, Bert-TiAttention, based on a shared Bert network and a fused additive attention mechanism, to improve the classification performance of the long-text author identification dataset Reuters_50_50. The long text is divided into two segments, and each segment is input into the Bert network with shared weights to obtain their respective representations. The attention mechanism is used to learn the correlation features between the two segments of text information, thereby assisting in learning the author’s writing style features and identifying the author. Compared with other commonly used author identification methods, this model can well capture the hidden writing style features in the text, and has better author identification model performance.