复杂文本多标签分类算法的设计与仿真  

Design and Simulation of Multi Label Classification Algorithm for Complex Text

作　　者：李楚贞 吴新玲 余育文 LI Chu-zhen;WU Xin-ling;YU Yu-wen(Institute of Information and Technology,Guangdong Polytechnic College,Zhaoqing Guangdong 526100,China;Institute of Computer Science,Guangdong Polytechnic Normal University,Guangzhou Guangdong 510665,China)

机构地区：[1]广东理工学院信息技术学院,广东肇庆526100 [2]广东技术师范大学计算机科学学院,广东广州510665

出　　处：《计算机仿真》2022年第5期299-303,共5页Computer Simulation

基　　金：广东省财政厅2019年教育政务管理学校德育项目“三全育人项目”(156028)。

摘　　要：由于已有算法未能通过卷积神经网络进行分类,导致分类结果不准确,分类复杂度提升,容错率下降。结合卷积神经网络,提出一种新的复杂文本多标签分类算法。首先在训练样本集中通过Bootstrap方法进行样本抽取,利用特征选择算法对抽取的特征进行评价。采用投票方法确定评价结果的特征权重,通过特征权重完成特征选择。然后,利用Word2vec工具将复杂文本特征提取结果转换为词向量,同时将句子整理为向量矩阵的形式。利用粒子群算法对卷积神经网络模型进行优化,进而实现复杂文本多标签分类。最终进行仿真测试,结果表明所提算法能够获取高精度的分类结果,降低分类复杂度,提升容错率。The traditional complex text multi-label algorithm has inaccurate classification results,high classification complexity and low fault tolerance.This paper presents a new multi-label classification algorithm for complex text.First of all,based on the Bootstrap method,samples were extracted from the training sample set.Secondly,a feature selection algorithm was introduced to evaluate the extracted features.Then,the voting method was applied to determine the characteristic weight of the evaluation results.Then,according to the result of feature weight,feature selection was completed.Then,the Word2 vec tool was used to convert the complex text feature extraction results into word vectors.Meanwhile,sentences were also converted into vector matrices.Then,a particle swarm optimization algorithm was applied to optimize the convolutional neural network model.Finally,complex text multi-label classification was completed.The results show that the algorithm has high precision classification results,low classification complexity and high fault tolerance.

关 键 词：卷积神经网络 复杂文本 多标签分类 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]