基于RCNN模型的英语文本摘要自适应分类方法  

An adaptive classification method of English text summarization based on RCNN model

作　　者：齐迹[1] QI Ji(Basic Courses,Shaanxi Railway Institute,Weinan 714000,Shaanxi Province,China)

机构地区：[1]陕西铁路工程职业技术学院基础课部,陕西渭南714000

出　　处：《信息技术》2023年第10期23-28,共6页Information Technology

基　　金：陕西铁路工程职业技术学院科研基金项目(KY2016-10)。

摘　　要：采用目前方法分类英语文本摘要时,存在精确度低、召回率低及文本分类时间较长的问题,为此,提出了基于RCNN模型的英语文本摘要自适应分类方法。利用期望交叉熵ECE和ECE′求值函数式计算词频权重并提取文本数据全部特征词融合组成特征词典。采用朴素贝叶斯方法选择最优主题,并通过Gibbs采样获取新文本特征,将新文本特征投入双循环神经网络中进行嵌入向量训练,通过词表示学习和多头注意池化两种结构实现英语文本摘要自适应分类。实验结果表明,所提方法精确度较高、召回率较高、文本分类时间较短。When using the current methods to classify English text abstracts,there are some problems,such as low accuracy,low recall and long text classification time.Therefore,an adaptive classification method of English text summarizations based on RCNN model is proposed.The expected cross entropy ECE and ECE’evaluation function are used to calculate the word frequency weight,and all feature words of text data are extracted to form a feature dictionary.The naive Bayesian method is used to select the optimal topic,and the new text features are obtained by Gibbs sampling.The new text features are put into the double loop neural network for embedded vector training.Adaptive classification of English text abstracts is realized through word representation learning and multi head attention pooling.Experiment results show that the proposed method has high accuracy,high recall and short text classification time.

关 键 词：RCNN模型 自适应分类 特征词典 词表示学习 多头注意力池化 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]