改进的图神经网络文本分类模型应用研究——以NSTL科技期刊文献分类为例

The Application of Improved Graph Convolutional Neural Network in Big Data Classification of Scientific and Technological Documents

[目的/意义]随着互联网数字资源的剧增,如何从海量数据中挖掘出有价值的信息成为数据挖掘领域研究的热点问题。文本大数据分类是这一领域的关键问题之一。随着深度学习的发展,使得基于深度学习的文本大数据分类成为可能。[方法/过程]针对近年来出现的图神经网络文本分类效率低的问题,提出改进的方法。利用文本、句子及关键词构建拓扑关系图和拓扑关系矩阵,利用马尔科夫链采样算法对每一层的节点进行采样,再利用多级降维方法实现特征降维,最后采用归纳式推理的方式实现文本分类。[结果/结论]为了测试该文所提方法的性能,利用常用的公用语料库和自行构建的NSTL科技期刊文献语料库对本文提出的方法进行实验,与当前常用的文本分类模型进行准确率和推理时间的比较。实验结果表明,所提出的方法可在保证文本及文献大数据分类准确率的前提下,有效提高分类的效率。

[Purpose/Significance]With the explosion of electronic data,big data mining has become a hot research issue in the field of data mining,and the classification of text big data based on deep learning is one of the key issues in this field.[Method/Process]In view of the low efficiency of text classification based on graph convolutional neural network in recent years,an improved method is proposed.This method uses text,sentences and keywords to construct topology diagram and relationship matrix,uses Markov chain sampling algorithm to sample nodes of each layer,and then uses multistage dimension reduction method to realize feature dimensionality reduction,and finally realizes text classification by inductive reasoning.[Result/Conclusion]The paper's methods,FSATGCN,GCN and other classification methods were tested by using the corpus constructed from NSTL journal literature.The results show that the model and method proposed in this paper can effectively improve the classification efficiency on the premise of ensuring the classification accuracy of big data in journal literature.