基于深度学习的文本特征提取研究综述  

Review of Text Feature Extraction Based on Deep Learning

作　　者：张千[1] 王庆玮 张悦 纪校锋 张宇翔 祝赫 赵昌志 ZHANG Qian;WANG Qing-wei;ZHANG Yue;JI Xiao-feng;ZHANG Yu-xiang;ZHU He;ZHAO Chang-zhi(School of Computer&Communication Engineering,China University of Petroleum(East China),Qingdao 266580,China)

机构地区：[1]中国石油大学(华东)计算机与通信工程学院

出　　处：《计算机技术与发展》2019年第12期61-65,共5页Computer Technology and Development

基　　金：中央高校基本科研业务专项基金(18CX02019A);科技部创新方法工作专项(2015IM010300)

摘　　要：文本特征项的选择是文本挖掘和信息检索的基础和重要内容。传统的特征提取方法需要手工制作的特征,而手工设计有效的特征是一个漫长的过程,但针对新的应用深度学习能够快速地从训练数据中获取新的有效特征表示。作为一种新的特征提取方法,深度学习在文本挖掘方面取得了一定的成果。深度学习与传统方法的主要区别在于,深度学习能自动地从大数据中学习特征而不是采用手工制作的特征,手工制作的特征主要依赖于设计者的先验知识,很难充分利用大数据;深度学习可以自动地从大数据中学习特征表示,并包括数以万计的参数。文中概述了用于文本特征提取的常用方法,并阐述了在文本特征提取及应用中常用的深度学习方法,以及深度学习在特征提取中的应用展望。The selection of text feature items is basic and important in text mining and information retrieval.Traditional feature extraction methods require hand-made features,and manual design of effective features is a long process.However,for new applications,deep learning can quickly obtain new and effective feature representation from training data.As a new feature extraction method,deep learning has made some achievements in text mining.The main difference between deep learning and traditional methods is deep learning can automatically learn features from large data rather than using hand-made features.Hand-made features mainly rely on designer's prior knowledge,which is difficult to fully use large data.Deep learning can automatically learn feature representation from large data and include tens of thousands of parameters.We summarize the common methods of text feature extraction and expound the deep learning methods commonly used in text feature extraction and application,as well as the application prospect of depth learning in feature extraction.

关 键 词：深度学习 特征提取 文本特征 自然语言处理 文本挖掘 

分 类 号：TP31[自动化与计算机技术—计算机软件与理论]