AM-BRNN:一种基于深度学习的文本摘要自动抽取模型  

AM-BRNN: Automatic Text Summarization Extraction Model Based on Deep Learning

作　　者：沈华东 彭敦陆[1] SHEN Hua-dong;PENG Dun-lu(School of Optical-Electrical and Computer Engineering, University of Shanghai for Science and Technology, Shanghai 200093, Chin)

机构地区：[1]上海理工大学光电信息与计算机工程学院,上海200093

出　　处：《小型微型计算机系统》2018年第6期1184-1189,共6页Journal of Chinese Computer Systems

基　　金：上海智能家居大规模物联共性技术工程中心项目(GCZX14014)资助;沪江基金研究基地专项项目(C14001)资助;国家自然科学基金项目(61003031)资助

摘　　要：是文本主要内容和核心思想的最小化表达,对从海量文本数据中快速寻找有价值的信息具有重要意义.利用深度神经网络Encoder-Decoder基本框架,通过引入注意力模型,提出文本摘要抽取的深层学习模型——AM-BRNN.论文先根据中文文本的语言特点,构建句子特征向量抽取算法,形成文本特征向量矩阵,再将其输入到AM-BRNN深层学习模型中,双向循环神经网络编码出中间语义向量,最后利用注意力模型与单向循环神经网络解码中间语义向量,实现摘要句子的抽取.实验结果表明,AM-BRNN能较准确且稳定的抽取摘要句子,相比其他模型具有更好抽取效果.Text summarization is the minimal expression of main contents and core idea of the documents. It is significant for seeking valuable information quickly from massive textual documents. Based on the fundamental framework of deep neural network EncoderDecoder and attention model,this paper proposes a textual summarization extraction model ——AM-BRNN which is based on deep learning model. According to the linguistic features of the Chinese documents,this paper develops an algorithm for extracting the feature＇s vectors from the documents to form the textual feature vector matrix. And then,the matrix is used as input to the deep learning model. The bi-directional recurrent neural network encodes the intermediate semantic vector. Finally,the attention model and the recurrent neural network are explored to decode the intermediate semantic vector to extract the abstract sentence. The experimental results show that the proposed algorithm,AM-BRNN can be used to extract abstract sentences more accurately and steadily as well as high performance compared with other existing algorithms.

关 键 词：文本摘要 深度学习 循环神经网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]