自然语言处理中的深度学习方法研究  

Research on deep learning in natural language processing

作　　者：陈静[1] 梁俊毅 CHEN Jing;LIANG Junyi(Guangxi Vocational&Technical Institute of Industry,Nanning 53000,China;Beihai Vocational College,Beihai,Guangxi 536000,China)

机构地区：[1]广西工业职业技术学院,南宁530001 [2]北海职业学院,广西北海536000

出　　处：《计算机应用文摘》2023年第17期130-132,136,共4页Chinese Journal of Computer Application

基　　金：广西工业职业技术学院2020年度科研项目:基于深度学习的“智慧课堂管家”系统研究与实现(桂工业院科)[2020]1号。

摘　　要：文本分类是自然语言处理中的一项基本任务。现有的大多数研究都依赖于显式或隐式的文本表示来解决这类问题,这些研究对较长的句子有较好的效果,但不能很好地应用于短文本分类。传统的多尺寸滤波器卷积神经网络能获得简单的词向量特征,但是容易忽略重要特征,因此文章提出了一种基于神经网络的短文本分类模型,该模型可以采用非线性滑动的方法与N-gram语言模型获得丰富的文本特征,并通过集中机制来挑选关键特征,然后采用池化操作来尽可能地保持文本特征的确定性。实验表明,与传统的卷积神经网络相比,该算法可以显著改善短文本分类的效果。Text classification is a basic task in natural language processing.Most existing research relies on explicit or implicit text representations to solve such problems,which work well for longer sentences but do not apply well to short text classification.The traditional multi-dimension filter convolutional neural network can obtain simple word vector features but easily ignore important features.Therefore,a short text classification model based on neural network is proposed in this paper.The model can obtain rich text features by using nonlinear sliding method and N-gram language model,and select key features by centralized mechanism.Pooling operations are then used to preserve as much deterministic textual features as possible.Experiments show that compared with the traditional convolutional neural network,the proposed algorithm can significantly improve the classification of short texts.

关 键 词：自然语言处理 情感分析 CNN 深度学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]