基于多元语义特征和图卷积神经网络的短文本分类模型  

A Short Text Classification Algorithm Combining Multivariate Semantic Features

作　　者：鲁富宇[1] 冷泳林[1] 崔洪霞 LU Fuyu;LENG Yonglin;CUI Hongxia(College of Information Science and Technology,Bohai University,Jinzhou 121003,Liaoning China)

机构地区：[1]渤海大学信息科学与技术学院,辽宁锦州121003

出　　处：《河南科学》2024年第5期625-630,共6页Henan Science

基　　金：国家自然科学基金资助项目(62006035);辽宁省教育厅重点攻关项目(JYTZD2023174);渤海大学研究生教育教学改革研究项目(YJG20230009)。

摘　　要：在互联网和社交媒体迅猛发展的背景下,网络中出现了大量的短文本数据,这些数据在舆情监控、情感分析和新闻分类等自然语言处理领域展现出了非常高的经济和学术价值.但短文本数据固有的特征给短文本分类带来了不小的挑战,如文本稀疏和缺乏丰富的上下文语义等.针对这些问题,提出了一种结合多元语义特征和图卷积神经网络(GCN)的短文本分类模型,该模型通过哈尔滨工业大学的语言技术平台获取短文本的多种语义特征,并将这些语义特征同短文本一起构建一个多元异构图,然后将其作为GCN的输入,利用GCN学习短文本更深层特征,最后通过Softmax函数获取每个类别的概率分布,进而实现短文本分类.试验结果表明,本模型在短文本分类的F1评分上比传统单一模型提高了4%.With the rapid development of the internet and social media,a large volume of short text data has surfaced online.Such data exhibit significant economic and academic value in natural language processing,including public opinion monitoring,sentiment analysis,and news categorization.However,inherent characteristics of short text data,such as text sparsity and the lack of rich contextual semantics,present significant challenges for short text classification.This paper proposes a short text classification model that integrates multiple semantic features with Graph Convolutional Networks(GCN)to tackle these challenges.This model acquires various semantic features of short texts from the Harbin Institute of Technology language technology platform and constructs a multi-element heterogeneous graph using these semantic features and short texts.Subsequently,it employs this graph as input for the GCN to delve into deeper features of short texts.Finally,the model utilizes the Softmax function to derive the probability distribution for each category,thus accomplishing short text classification.Experimental results demonstrate that this model enhances the F1 score in short text classification by 4%compared to traditional single models.

关 键 词：短文本 多元异构图 语义特征 图卷积神经网络 分类模型 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]