Research On Automatic Evaluation Model And Algorithm Based On MOOC Video Subtitles And Learning Data

MOOC is a kind of online course that has emerged and developed very rapidly in recent years.By sharing educational resources to everyone in the form of video,blog and document on the Internet,it makes learning easier and more convenient.Hence,MOOC has made a great impact on higher education.The large-scale increase of MOOC learners makes the automation of evaluation system more and more important.Therefore,we carry out a series of researches on how to automatically generate questions and automatically compose test papers in this paper.Our work and innovations mainly include:· We innovatively propose an algorithm of automatically generating questions from MOOCs video subtitles.Firstly,we use wiki knowledge base to extract facts of interest and visualize them in the form of knowledge graph.Secondly,we propose a novel template-based method to generate questions from knowledge graph.The experimental results based on the public dataset Simple Questions show that our proposed template-based method outperforms other competitive methods by 6.99% in terms of BLEU.· We innovatively propose an algorithm of automatically composing test papers with the support of the learners' usual learning data.We firstly determine the overall difficulty of test paper according to learners' usual learning data,and then use genetic algorithm to generate test paper considering multiple constraints.The experimental results based on a MOOC course show that the mean absolute error of prediction model is roughly around 10 points on 100 points scale and we can successfully achieve the intelligent composition of test papers with various objectives optimized.In summary,we propose and implement an automatic evaluation model based on MOOCs video subtitles and learners' learning data,which contributes to the automation of MOOCs' evaluation system.