外部知识与内部上下文语义聚合的短文本新闻虚假检测模型

打开文本图片集

中图分类号：TP391 文献标志码：A

Short Text News Fake Detection Model Based on Aggregating External Knowledge and Internal Contextual Semantics

QIU Yanfang 1 ， ZHAO Zhenyu 2 ， SUN Zhijie'， MA Kun’ ， JI Ke1 ， CHEN Zhenxiang 1

（1.a.School of Information Science and Engineering，b.Shandong KeyLaboratory of Ubiquitous Inteligent Computing， University of Jinan，Jinan 250O22，Shandong，China; 2. Shandong Talent Development Group Information Technology Co.， Ltd.， Jinan ， Shandong，China）

Abstract：To adressthe problem ofsemantic feature sparsityin shorttext news and the neglectof the homology between external knowledge and thesemanticsof short-text news，ashort text news fake detection model basedonagregating external knowledge and internal contextual semantics （EKCS-ST）was proposed.A news feature information network was constructed，which included three typesof external knowledge，such as news topics，authors，and entities，to enrichthe semantic featuresof short text news.The exteral knowledge graph features of the news were generated through graph convolution.The newstext was fed intoa text encoder to capture internal contextual semantic features.These external knowledgegraph featuresand internal contextual semantic features were thenused in a context-aware computation to strengthen thecorrelation between external knowledgeand contextual semantics.Theatention mechanism wasutilized to selectand enhance the keyfeaturesof the news，whiletheloss errorfor minority-classnews was increased to mitigate the data imbalance issue.The results show that F1 score of the proposed model，which is the harmonic mean of precision and recall，is O.86，outperforming BERT and TextGCN models by 18% and 17% ，respectively，validating the effectiveness of the model.

Keywords： short text news fake detection；external knowledge；attention mechanism；semantic feature

自媒体发布了大量快讯、头条等表达简短扼要的短文本新闻，未经鉴别的新闻真实性无法保证[1]。（剩余13876字）