基于大语言模型两阶段微调方式的高校心理测评报告信息提取研究

打开文本图片集

中图分类号：TP399 文献标识码：A 文章编号：1006-8228（2025）11-41-05

Research on Information Extraction from University Psychological Assessment Reports Using a Two-Stage Fine-Tuning Approach Based on Large Language Models

Li Ziying，Hu Zhichen， Zhao Yongjie （Zhenjiang College，Zhenjiang，Jiangsu ，China）

Abstract：Inthegeneralpsychologicalassessmentofcollgestudents，psychologicalasessmentreportsreflectindividuals' psychologicalstatesthroughkeyinformationsuchassymptomdescriptions，emotionalexpressionsandinterventionrecords. However，suchreportshaveproblemssuchasalargenumberofprofessonalterms，significantvariabilityinexpressionsanda smallamountoflabeleddatawhichresultinpoorperformanceoftraditionalinformationextractionmethods.Thispaperpresentsa two-stagefine-tuningmethodwiththehelpoflargelanguagemodels，namedPsych-LLME.First，aninstructiontemplatefor psychologicaldiagnosisandtreatmentiscreatedandsimulateddataisgeneratedOnthisbasis，dataaugmentationtechiquesare usedtoexpandtheavailabletrainingsamples.Ten，thelow-rankadaptationtechnologyisusedtoquicklyfine-tunelargelanguage modelssuchasLlama2toimprovetheirunderstandingofpsychologicalprofesionaltexts.Thisresearchprovidesaneffective solutionfortheintelligentprocessingofcollegepsychoogicalasessmentreportsandisofgreatsignificanceforpromotingthe digital transformation of college mental health services.

Keywords：LargeLanguageModels;sychologicalAssessmentReports;UniversityMentalHealthServices;DigitalTransfoation

0引言

心理测评报告作为心理健康评价的关键载体，不仅记录了个体的症状表现、情绪特点和认知模式，还贯穿于干预进程的始终，为临床诊断、疗效跟踪和心理研究提供了重要依据1。（剩余6605字）