一种基于混合神经网络的抽取式文本摘要方法  

On Extractive Summarization Via Hybrid Neural Networks

作　　者：林晶[1] 

机构地区：[1]怀化学院计算机科学与工程学院,湖南怀化418008

出　　处：《怀化学院学报》2017年第5期71-74,共4页Journal of Huaihua University

基　　金：湖南省重点实验室项目"武陵山区生态农业农情摘要关键技术研究"(No.ZNKZ2014-8)

摘　　要：能够帮助人们快速浏览和理解文档或文档集,传统抽取式摘要方法高度依赖于人工特征,本文设计了一种基于混合神经网络(FNN)的文档摘要方法,它包含了一个分层文档编码器及一个基于关注的抽取器,能够不依赖于人工特征自动抽取句子产生摘要.实验结果表明,该方法效果较好.A document or a set of documents are easy for readers to read and understand fast by their summaries.But traditional extractive summarization relies heavily on human-engineered features. Hence in this work a summarization approach was proposed based on hybrid neural networks（FNN）. Our model includes a neural network-based hierarchical document reader or encoder and an attention-based content extractor. The proposed approach can automatically generate summary by extracting salient sentences from documents. Experiments show that our model outperforms previous state-ofthe-art methods.

关 键 词：文本摘要 句子抽取 混合神经网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]