基于双向编码器表示和图卷积网络的文本多重语义检测研究  

Research on Textual Multiple Semantic Detection Based on Bidirectional Encoder Representations and Graph Convolutional Networks

作　　者：秦振凯 朱雨燕 沈洁 陈悦 魏文婧 QIN Zhenkai;ZHU Yuyan;SHEN Jie;CHEN Yue;WEI Wenjing(School of Information Technology,Guangxi Police College,Nanning Guangxi 530028;School of Criminal Science and Technology,Guangxi Police College,Nanning Guangxi 530028)

机构地区：[1]广西警察学院信息技术学院,广西南宁530028 [2]广西警察学院刑事科学与技术学院,广西南宁530028

出　　处：《软件》2025年第3期5-7,共3页Software

基　　金：国家级大学生创新创业训练项目(202413520003);广西高校中青年教师科研基础能力提升项目(2024KY0902);广西哲学社会科学规划项目(24TQF007);2024年中国—东盟(广西)警学研究院研究课题(DMJXZD202401)。

摘　　要：随着互联网的快速发展,海量的信息需要监管部门进行分析和处理。然而,文本内容时常存在多重语义,这对现有的识别方法提出了挑战。虽然现有方法在文本筛选和语义检测方面已取得一定成效,但其在识别复杂语义和细微差异方面仍有提升空间。为解决监管过程中存在的难题,本文对双向编码器表示(BERT)和图卷积网络(GCN)联合检测文本多重语义的能力进行研究,提出了一种BERT-GCN架构。该架构以文本表示作为输入,学习文本中的复杂结构和语义模式,从而检测文本内容的不同含义。实验结果表明,BERT-GCN的检测效率高于其他基线方法。With the rapid development of the Internet,a huge amount of information needs to be analyzed and processed by regulators.However,multiple semantics exist in text content from time to time,which poses a challenge to existing recognition methods.Although existing methods have achieved some success in text screening and semantic detection,they still have room for improvement in recognizing complex semantics and subtle differences.In order to address the challenges in the regulatory process,this paper investigates the ability of bidirectional encoder representation(BERT)and graph convolutional network(GCN)to jointly detect multiple semantics in text,and proposes a BERT-GCN architecture.This architecture takes text representation as input,learns complex structures and semantic patterns in the text,and detects different meanings of the text content.The experimental results indicate that the detection efficiency of BERT-GCN is higher than other baseline methods.

关 键 词：文本检测 图卷积网络 BERT TRANSFORMER 词嵌入 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]