基于多模态数据的精准在线测试模型  

An Accurate Online Testing Model Based on Multimodal Data

作　　者：陈波[1] 陆天易 于泠[2] 刘惠林 CHEN Bo;LU Tian-yi;YU Ling;LIU Hui-lin(School of Artificial Intelligence,Nanjing Normal University,Nanjing,Jiangsu,China 210023;School of Mathematical Sciences,Nanjing Normal University,Nanjing,Jiangsu,China 210023)

机构地区：[1]南京师范大学人工智能学院,江苏南京210023 [2]南京师范大学数学科学学院,江苏南京210023

出　　处：《现代教育技术》2023年第4期92-100,共9页Modern Educational Technology

基　　金：江苏省“十四五”教育科学规划重大课题“未来学校建设研究”(项目编号:A/2021/05);江苏省高等教育学会重点资助课题“在线教学质量评价体系研究”(项目编号:2021-Z07)资助。

摘　　要：随着“互联网+”时代的到来,在线测试开始受到广泛的关注,并逐渐被应用于学习系统中。然而,目前常见的在线测试系统多以学习者的做题成绩、做题时间等单模态数据为依据计算测试成绩,对测试过程的感知和反馈很少,导致在线测试成绩的客观性和真实性不足。为此,文章结合多模态数据的分析特点,提出了一种基于多模态数据的精准在线测试模型,重点研究了眼动和键鼠行为数据作为评价数据源的应用方法及其有效性。实验结果表明:眼动和键鼠等多模态数据的融入有助于客观、真实地反映学习者的测试过程;模型能够有效提高在线测试结果的准确性,多级融合方法也使多模态数据具有较好的可解释性;同时,非侵入性的数据获取使模型具有较高的实用性。文章对多模态学习行为数据获取、挖掘、融合与应用的研究,旨在为在线学习评价提供参考,并为学习平台客观、全面把握学习者学习成效和实现个性化学习服务提供有效支撑。With the arrival of the“Internet+”era,online testing has begun to receive widespread attention and gradually been applied to learning systems.However,the current common online testing systems are mosty based on the single-modal data such as learners’test scores and test-taking time to calculate test scores,which resulted in little perception and feedback on the testing process and led to insufficient objectivity and authenticity of online testing results.Therefore,combining with the analysis characteristics of multimodal data,this paper proposed an accurate online testing model based on multimodal data,and gave the selected multimodal data and an evaluation index system designed from three dimensions of knowledge mastery degree,thinking activity degree,and cognitive input degree,as well as a multi-level fusion algorithm.Meanwhile,the application methods and effectiveness of behavior data of eye movement and keyboard-mouse as evaluation data sources were mainly studied.Experimental results indicated that the integration of multimodal data,such as eye movement and keyboard-mouse helped to objectively and truly reflect learners’testing process,and the model could improve the accuracy degree of online testing results.Meanwhile,the multilevel fusion method made the multimodal data more interpretable,and the acquisition of non-intrusive data made the model more practical.The acquisition,mining,integration and application of multi-modal learning behavior data in this paper were expected to provide reference for the online learning evaluation,and offer effective support for learning platforms to objectivly and comprehensivly understand learners’learning outcomes and achieve personalized learning services.

关 键 词：在线测试 多模态数据 行为分析 眼动检测 

分 类 号：G40-057[文化科学—教育学原理]