基于系统性文献综述的多模态学习分析研究进展与前瞻  

Research Progress and Prospects of Multimodal Learning Analytics Based on Systematic Literature Review

作　　者：尹睿[1] 何淑茵 YIN Rui;HE Shuyin

机构地区：[1]华南师范大学教育信息技术学院,广东广州510631

出　　处：《现代远程教育研究》2022年第6期54-63,共10页Modern Distance Education Research

基　　金：2019年广东省高等教育教学改革项目“设计思维导向的师范生信息技术应用能力在线开放课程混合式教学改革与实践创新”;华南师范大学“挑战杯”金种子培育项目“融合多源数据分析的高校混合学习投入评价模型研究”。

摘　　要：多模态学习分析是指利用“多模态”的思想和方法对学习者的内在学习状态、特征与变化进行深度诠释,旨在挖掘学习规律、优化学习过程、促进精准教学。在教育大数据研究范式下,多模态学习分析应运而生并形成了诸多研究成果。运用系统性文献综述方法对国内外多模态学习分析相关研究进行梳理发现:当前多模态学习分析研究聚焦于数据采集、数据融合和数据建模三大方面。在数据采集上,得益于智能感知技术的发展,文本、语音、动作、表情、眼动、生理等模态数据备受关注且获取方式愈加便捷;在数据融合上,主要以数据层融合、特征层融合、决策层融合为主,因应深度学习算法的发展,混合式融合初见端倪;在数据建模上,涌现出面向知识、认知、情感、交互状态的学习者模型,且基于多元学习状态的整体性模型日益受到重视。未来多模态学习分析研究应加强情境感知,实现场景数据的混合采集;深挖理论基础,促进数据融合的科学精准;重视情境依存,强化数据建模的情境适用。Multimodal learning analytics aims to use ideas and methods of“multi-modality”to deeply interpret learners’internal learning state,characteristics and changes,so as to explore the learning law,optimize learning processes and promote precision teaching.Under the educational big data research paradigm,multimodal learning analytics has emerged,which results in many relevant research products.An analysis of the foreign and domestic relevant research on multimodal learning analytics with a systematic literature review method found that the current multimodal learning analysis research focuses on data acquisition,data fusion and data modeling.In terms of data acquisition,benefiting from the development of artificial awareness technology,multimodal data such as text,audio,motion,expression,eye movement and physiology,has attracted much attention and has been more easily captured.As to data fusion,it mainly focuses on data-level,feature-level and decision-level fusion,and blended-level fusion emerges with the development of deep learning algorithms.As for data modeling,learner models for knowledge,cognition,emotion and interaction have emerged,and the holistic model based on the multiple learning state has received increasing attention.In the future,multimodal learning analytics should strengthen the context-awareness to realize data acquisition in blended scenarios,dig deep into theoretical foundations to promote scientific and accurate data fusion,attach importance to contextual dependence to strengthen the situational applicability of data modeling.

关 键 词：多模态学习分析 数据融合 数据建模 学习者模型 系统性文献综述 

分 类 号：G434[文化科学—教育学]