摘要
会话辅导系统是智能辅导系统的一种特殊形态或拓展形式,主要通过模仿人类导师进行个别化、适应性的教学对话,实现个性化教学功能。受制于技术发展水平,传统会话辅导系统存在一些根本性缺陷,如缺乏通用性与灵活性、开发成本高昂、推广困难、准确性不足等。大模型为会话辅导系统的升级带来了巨大机遇,将引发会话辅导系统的开发范式转换:开发目标由原先的“排演对话教学片断”升级为“自主开展完整对话教学”,开发逻辑由以“预设”与“匹配”为核心转换为以“微调”与“链接”为核心。为此,会话辅导系统由原来的内外双循环架构,拓展为外循环—中循环—内循环的三重嵌套循环架构模式,以实现学习者对话片段、学习任务完成过程,以及课程学习历程的完美融合。而针对系统架构中关键且存在技术瓶颈的教学设计能力、对话设计与生成能力,可以通过学习任务分析路由链、教学设计顺序链、教学对话路由链等复杂的提示语工程,以及使用优质的元认知标注数据、对话教学案例标注数据以及历史数据+RLHF技术微调大模型等办法加以训练,以此提高会话辅导系统开展对话教学的质量和效果。
Conversational tutoring system is a special type or an extended form of the intelligent tutoring system,which achieves personalized teaching functions mainly by imitating human tutors for individualized and adaptive teaching dialogues.Due to the level of technological development,the traditional conversational tutoring system has some fundamental defects,such as the lack of versatility and flexibility,high development costs,difficulty in promotion,and lack of accuracy.The large model brings a huge opportunity for the upgrade of the conversational tutoring system and will lead to its development paradigm shift:the development goal is upgraded from the original“rehearsing dialogue teaching fragments”to“independently carrying out complete dialogue teaching”,and the development logic is changed from“preset”and“matching”as the core to“fine-tuning”and“linking”as the core.To this end,the conversational tutoring system has been expanded from the original internal and external dual cycle architecture to the triple nested cycle architecture mode of outer loop,middle loop and inner loop,so as to realize the perfect integration of learners’dialogue fragments,learning task completion process,and course learning process.The teaching design capabilities,and dialogue design and generation capabilities,which are crucial and have technical bottlenecks in the system architecture,can be trained through complex prompt engineering such as learning task analysis routing chain,instructional design sequence chain,teaching dialogue routing chain,and the use of high-quality metacognitive annotation data,dialogic teaching annotation data,and historical data+RLHF technology to fine-tune large models,so as to improve the quality and effect of dialogic teaching by the conversational tutoring system.
作者
刘华
祝智庭
LIU Hua;ZHU Zhiting
出处
《现代远程教育研究》
CSSCI
北大核心
2024年第3期11-19,共9页
Modern Distance Education Research
基金
2018年度国家社会科学基金重大项目“信息化促进新时代基础教育公平的研究”(18ZDA335)
2023年度江苏省教育科学规划重点项目“教育本位的人机协同教学研究”(B/2-23/01/122)。
关键词
会话辅导系统
大语言模型
多模态大模型
对话教学
Conversational Tutoring System
Large Language Model
Multimodal Large Models
Dialogic Teaching
