A deep ensemble learning model for Chinese spelling check

Yaoyao Wu, Ruizhang Huang, Lina Ren, Ruina Bai,
A deep ensemble learning model for Chinese spelling check,
Engineering Applications of Artificial Intelligence,
Volume 162, Part D,
2025,
112469,
ISSN 0952-1976,
https://doi.org/10.1016/j.engappai.2025.112469.
(https://www.sciencedirect.com/science/article/pii/S095219762502500X)
Abstract: Existing Chinese Spelling Check models are individual end-to-end models with different tendencies, which cannot fully cover all kinds of spelling errors and achieves advanced performance in every aspect. In this paper, a deep ensemble learning model for Chinese Spelling Check is designed, which captures the correct Chinese spelling answers proposed by most of the candidate ckeck models. Considering the phonological, visual and semantic characteristics of all candidate answers, it can even discover new correct answer which is not found by any of the candidate ckeck models. Specifically, in order to learn a hybrid representation of each input correction answer provided by each candidate ckeck model that includes phonological, visual, and semantic characteristics, a hybrid representation learner is designed and does not need to consider the compatibility of the candidate ckeck models. A deep ensemble correction network is designed to integrate all the hybrid representations and finds a final correction answer that considers all the useful information of all input correction answers. Moreover, based on the deep ensemble correction network, the deep ensemble learning model can be easily extended to involve new candidate ckeck models. Experimental results on the benchmark demonstrate that our proposed model largely outperforms any of its aggregated individual ckeck models.
Keywords: Deep ensemble learning; Chinese spelling check; Hybrid representation learning; Semantic fusion; Extension of ensemble