基于Siamese LSTM的中文多文档自动文摘模型

在文本信息数量迅速增长的环境下,为提升阅读效率,提出一种基于深度学习的多文档自动文本摘要模型。在传统文摘模型的基础上将Siamese LSTM深度学习网络应用到文本相似度计算中,计算曼哈顿距离来表征文本相似度,并采用去除停用词的方法改进该网络模型以提升计算效率。实验结果表明,使用Siamese LSTM与传统余弦相似度等方法相比,生成的文摘在语义方面更贴近主题,质量更高,整个文摘系统的工作效率也显著提升。

CHINESE MULTI-DOCUMENT AUTOMATIC SUMMARIZATION MODEL BASED ON SIAMESE LSTM

With the rapid growth of text information, a multi-document automatic text summarization model is proposed to improve reading efficiency. On the basis of traditional abstract model, the Siamese LSTM deep learning network was applied to calculate text similarity and the Manhattan distance was calculated to represent the text similarity. Besides, the network model was improved by removing the stop words to improve the computing efficiency. The experimental results show that compared with the traditional method of calculating the cosine value, the generated summarization is closer to the subject in terms of semantics, and the quality is higher. The efficiency of the summarization system has also been improved significantly.