大语言模型驱动的多模态实验报告自动批改

打开文本图片集

中图分类号：TP391.4；TP181 文献标识码：A 文章编号：2096-4706（2025）12-0079-06

Automatic Correction of Multimodal Experimental Reports Driven by LLM

XU Jining， HUANG Nan， SONG Hao （SchoolofElectricalandControlEngineering，North China UniversityofTechnology，Beijing10o144，China）

Abstract：Automatic correction of experimental reports isan important task in the field of intellgent education. FollowingteOBEconcept，tepaperreproduces thteacher'sorrctionideatransfoms tesoingitesitoquestons，and coordinates temultimodalresposeiformatiosuchastext，tables，ndpictures tosoresostoblosetoteactuateaching and curiculum construction needs.Intheunderstanding and scoring stageof multimodal information，on the basisof Deep Learming，LLMisitroduced torealizethecontentextractionandtransformationoftablequestions，andsolvethdifculties of positioningandlogicaldiscrimination.Fortextcontent，BERTisusedtounderstand.Fortheimagecontent，theself-training modelconstructed by thecombination ofBERTand ResNet-18 isused to scale the image matching weightsforthe image featureevaluationingraphicquestions.Theschemeusessmallsampledata fortraining，adapts todiferent subject experiments， andovercomes the pain points such as insuffcient generalization and migration caused byrelying onalarge amount of data training.Through the correction test of two courses，the average accuracy of the report score reaches 92.20% ，bridging the gap of automatic correction of non-customized experimental reports.

Keywords： automatic correction of experimental report; Deep Learning; LLM

0 引言

随着人工智能技术融入教育，各类作业考试的自动评分系统使教师的工作变得更加高效。（剩余7185字）