Title: Chinese machine reading comprehension based on deep learning neural network

Authors: Chao Ma; Jing An; Jing Xu; Binchen Xu; Luyuan Xu; Xiang-En Bai

Addresses: School of Electrical and Electronic Engineering, Shanghai Institute of Technology, Shanghai, 201418, China ' School of Electrical and Electronic Engineering, Shanghai Institute of Technology, Shanghai, 201418, China ' China Baowu Design Institute, Baosteel Engineering & Technology Group Co., Ltd, Shanghai, China ' Department of Control Science and Engineering, Tongji University, Shanghai, 201804, China ' Department of Control Science and Engineering, Tongji University, Shanghai, 201804, China ' School of Electrical and Electronic Engineering, Shanghai Institute of Technology, Shanghai, 201418, China

Abstract: Most of the existing machine reading comprehension (MRC) models mainly learn the attention mechanism of interactive alignment between questions and texts, and there are still great difficulties in the characterisation of ultra-long texts. The complex grammar, diverse expressions, and flexible structure of the Chinese corpus make the MRC task even more difficult. In this paper, a transformer-based neural network model extremely brisk and clever net (XBCNet) was proposed for the MRC problem of Chinese text. XBCNet expands the receptive field by introducing a stack of dilated convolution neural networks, and it is also closely coupled with the word vector algorithm. This process aims to reduce the training and inference time of the model, while alleviating the drawback of long questions that are difficult to answer accurately. The experimental results demonstrate that XBCNet improves the performance of MRC under general texts and obtains the best inference effect in limited computing resources.

Keywords: natural language processing; NLP; Chinese machine reading comprehension; deep learning; neural networks.

DOI: 10.1504/IJBIC.2023.131888

International Journal of Bio-Inspired Computation, 2023 Vol.21 No.3, pp.137 - 147

Received: 16 Mar 2022
Accepted: 26 Dec 2022
Published online: 04 Jul 2023 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article