基于语义图增强注意力网络的症状属性分类方法

打开文本图片集

中图分类号：TP391 文献标志码：A 文章编号：1001-3695（2026）01-006-0053-07

doi：10.19734/j.issn.1001-3695.2025.06.0203

Symptom attribute classification method based on semantic graph-enhanced attention network

JiaHeming，LiWei，LiBo，ZhangZhidong† （StateKeLabratoryfExtreeEiroentOptoelectronicDyamicMeasurementTecoloy&Istruent，NorthUieitfina， Taiyuan 030003，China）

Abstract：Symptomatributeclassificationinmedicaldialoguesplaysacriticalroleinautomaticdiagnosissystems.Thetask aims toientifytheatributecategoriesof symptoms describedindialogue texts.However，existingapproachesoftenstrugleto modellong textsandfailtocapturesuicientsemanticdependencies，which limits theirperformance，especiallyonminority classes.Toaddressthesechallnges，thispaperproposedelation-awaregaphattetionnetwork forsymptomatributeclassification.The proposed method integratedsymptom-centered textsegmentation，afusedencoding strategy，andadependencybasedrelational graphatentionmechanism toenhancecontextualrepresentationsatmultiple levels.This paperevaluatedthe approach on the CHIP-MDCFNPC dataset. And the proposed method achieves an F1 score of 72.13% and a macro- F1 of （204 77.94% ，outperforming baseline models by 1.76% and 1.77% ，respectively. The proposed method effectively improves symptomatributeclassificationinlong medical dialoguesanddemonstrates particularlystrong performanceonminorityclasses， providing valuable insights for developing reliable automatic diagnostic systems.

KeyWords：symptom attribute classification；text segmentation；relational graph attention mechanism

0 引言

随着人工智能技术的发展，基于任务型对话的自动诊断系统在医疗辅助诊断中展现出广泛前景。（剩余17937字）