融入隐式情感和主题增强分布的网络敏感信息深度识别研究  

Deep Recognition of Network Sensitive Information by Integrating Implicit Emotion and Topic Enhanced Distribution

作　　者：吴树芳[1] 尹凯 吴汭漩 朱杰[3] WU Shufang;YIN Kai;WU Ruixuan;ZHU Jie(School of Management,Hebei University,Baoding 071000,China;School of Foreign Studies,Hebei University,Baoding 071000,China;College of Mathematics and Information Science,Hebei University,Baoding 071000,China)

机构地区：[1]河北大学管理学院,河北保定071000 [2]河北大学外国语学院,河北保定071000 [3]河北大学数学与信息科学学院,河北保定071000

出　　处：《情报科学》2024年第5期111-119,共9页Information Science

基　　金：河北省人文社会科学研究重大课题攻关项目“基于大数据的河北省网络治理机制研究”(ZD202102)。

摘　　要：【目的/意义】网络敏感信息是网络生态治理的重要对象,对其进行精准识别可以有效推动网络空间净化的进程。【方法/过程】为提高网络敏感信息识别性能,提出一种融入隐式情感和主题增强分布的深度识别方法:首先利用提出的IME方法度量敏感词的隐含情感,并将其与信息的原始情感融合获取网络信息的隐式情感特征;然后结合网络敏感信息的特殊性改进BTM模型,并基于改进模型获得网络信息的敏感主题增强分布;最后利用深度学习中的注意力机制将隐式情感特征、敏感主题特征与传统的敏感词特征、语义特征融合,实现对网络敏感信息的识别。【结果/结论】实验结果表明,本文挖掘的隐式情感特征和敏感主题特征均能有效提升网络敏感信息的识别性能,与已有方法相比,多特征融合方法在进行网络敏感信息识别时性能较优。【创新/局限】深入挖掘敏感信息的隐式情感特征和敏感主题特征,融合多种特征实现网络敏感信息的有效识别,为网络生态治理提供理论支撑,但信息中敏感主题的增强分布有待进一步研究。[Purpose/significance]Network sensitive information is an important object of network ecological governance.It can effec-tively promote the process of cyberspace purification by accurately identifying sensitive information on the network.[Method/process]In order to improve the recognition performance of network sensitive information,this study proposes a deep recognition method for network sensitive information that integrates implicit emotion and topic enhanced distribution.First,this study utilizes the proposed IME method to measure the implicit emotion of sensitive words.The implicit emotion feature of network information is obtained by fus-ing the implicit emotion of sensitive words with the original emotion of the information.Then,this study improves the BTM model by combining the special characteristics of network sensitive information,and extracts the sensitive topic enhanced distribution of net-work information based on the improved model.Finally,this study adopts the attention mechanism to fuse implicit emotion feature and sensitive topic feature with traditional sensitive word feature and semantic feature to realize the deep recognition of network sensitive information.[Result/conclusion]The experimental results show that both implicit emotion feature and sensitive topic feature of net-work information can effectively improve the sensitive information recognition performance.Compared with the existing methods,the network sensitive information recognition method based on multi feature fusion has better performance.[Innovation/limitation]This study deeply explores the implicit emotion feature and sensitive topic feature of sensitive information,and integrates multiple features to achieve effective recognition of network sensitive information.This study provides theoretical support for network ecological gover-nance,but further research is needed to enhance the distribution of sensitive topics in information.

关 键 词：敏感信息 隐式情感 主题增强分布 敏感主题 深度识别 

分 类 号：G253[文化科学—图书馆学]