基于长短期记忆网络的英语标题自动生成  

Automatic Generation of English Titles Based on Short-term and Long-term Memory Networks

作　　者：千颖利[1] QIAN Ying-li(Xianyang Vocational&Technical College,Xianyang 712046 China)

机构地区：[1]咸阳职业技术学院,陕西咸阳712046

出　　处：《自动化技术与应用》2024年第4期71-73,共3页Techniques of Automation and Applications

基　　金：陕西省高校学生工作研究课题(2022XKT81);咸阳职业技术学院2022年度教学改革课题(2022JYB02)。

摘　　要：为实现英文文本标题的自动化生成,研究一套基于长短期记忆网络的句子级LSTM编码策略,并在标题生成模型中引入注意力机制来获取英文文本的上下文向量,保留文本中的重要信息。在此基础上,通过负对数似然函数来对模型加以训练。最后通过Byte Cup 2018数据集对本文提出的英语标题自动生成算法进行实验,并通过过ROUGE-N指标对标题生成质量加以评价。实验研究发现,所提出的句子级LSTM编码方案在英文文本标题生成准确性方面相比于其他常规摘要生成模型来说具有显著优势。In order to realize the automatic generation of English text titles,a sentence level LSTM coding strategy based on long-term and short-term memory networks is studied,and attention mechanism is introduced into the title generation model to obtain the context vector of English text and retain the important information in the text.On this basis,the model is trained by negative log likelihood function.Finally,the automatic title generation algorithm proposed in this paper is tested by Byte Cup 2018 data set,and the title generation quality is evaluated by rouge-n index.The experimental results show that the sentence level LSTM coding scheme proposed in this paper has significant advantages over other conventional abstract generation models in terms of the accuracy of English text title generation.

关 键 词：长短期记忆网络 英语文本 标题自动生成 句子级LSTM编码 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TP391.43[自动化与计算机技术—控制科学与工程]