融合混合提示與位置感知的突發(fā)事件抽取模型

打開文本圖片集

Contextual enhancement template for emergency event extraction via position-aware attention mechanism

Guo Jialiangla，1b，Wang Weila，1b，2?，JuJingla，1b，WangYafeila，b （1.a.Scholfftio&lecclEinebebotouritrotectofas&s beiUniverstydnbeOoiclooU ty ，Chengdu 610500，China）

Abstract：Thetask ofemergency event extraction aims todetectand extracttheeventtypes andelementscontained within newsreportsof emergency events，providingdetailedand structured information forpublicsafetyand emergencyresponse.To addresstheissueofnglectingcontextualinformationcausedbysparsetextualdatainthedomainofemergencyeventextraction，this paper proposedanemergency event extraction model basedonacontext-enhanced hybridprompttemplateand a position-awareatentionmechanism.Firstly，itdesignedacontext-ancedhybridprompttemplate，hichincorporatedtaskrelatedpromptsintotheinputtexttoimprovethemodel’sunderstandingandreasoningabilityforthetask.Additionaly，itintroducedaposition-aware atention mechanism，capturingkeysemanticinformationfromboth forwardand backwarddirections toovercomethesymmetrylimitationsof traditionalattentionmechanisms.ExperimentalresultsonaChineseemergencyevent corpus show that the proposed method achieves F1 scores of 91. 8% for event type detection and 82.0% for event element extraction.

Keywords：emergency events；event extraction；event element identification；prompt learning；atention mechanism

0引言

隨著互聯(lián)網(wǎng)和移動通信技術(shù)的迅速發(fā)展，社交媒體平臺、新聞網(wǎng)站和公眾信息發(fā)布系統(tǒng)成為突發(fā)事件信息傳播的重要渠道。（剩余19404字）