Title: Automatic generation of Chinese abstract based on vocabulary and LSTM neural network

Authors: Guijun Zhang

Addresses: Information Institute, Shanxi Finance and Taxation College, Taiyuan, Shanxi, China

Abstract: Most methods of Chinese short text summarisation are based on extraction, and it's hard to guarantee that the abstract is consistent. In this paper, we present an effective automatic method of Chinese abstract by using vocabulary and long-short term memory neural networks. The method utilises the seq2seq architecture, and introduces the candidate vocabulary in the decoding stage, to reduce the decoder vocabulary size. Thus, the training process is faster and the result is more concise and grammatical. In the end, experimental results validate the correctness and effectiveness of the method by taking a Large-Scale Chinese Short Text Summarisation (LCSTS) data set and Recall-Oriented Understudy for Gisting Evaluation (ROUGE).

Keywords: Chinese text summarisation; Seq2Seq model; LSTM neural network.

DOI: 10.1504/IJWMC.2020.10033260

International Journal of Wireless and Mobile Computing, 2020 Vol.19 No.3, pp.241 - 248

Received: 15 May 2019
Accepted: 19 Dec 2019

Published online: 30 Oct 2020 *

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