Research On Key Issues Of MOOC-Oriented Personalized Learning Path Recommendation

Massive open online courses(MOOC)make high-quality educational resources accessible online via Internet,where the resources can be followed and learned by the majority of learners,leading to a new teaching mode.In the past decade of development,the market scale of MOOC continues to rise.Meanwhile,due to the influence of COVID-19,MOOC has been an indispensable teaching tool.With the continuous rise of courses and learners in MOOC platforms,the current monotonous interaction and feedback are not enough to cover large-scale and diversified users.The research and application of intelligent MOOC teaching mode is imperative.In the intelligent MOOC teaching mode,the tutoring system based on artificial intelligence technology is embedded into the MOOC platforms,aimed at providing learning strategies for diversified learners based on the existing courses.However,due to the lack of systematic and personalized user modeling,knowledge concept-level course management,and personalized learning path recommendation,intelligent tutoring system in MOOC has not yet formed a feasible solution.To solve the above problems,this paper focuses on the key issues of MOOC-oriented personalized learning path recommendation,and carries out exploratory researches from three aspect,that is,user modeling,knowledge concept graph,and learning path recommendation.Firstly,for the task on design and implementation of user modeling in MOOC,this paper proposes a multidimensional user modeling system covering static,dynamic,and opinion features,based on the existing studies on user modeling in the field of education.Further,this paper carries out practical researches in view of the lack or deficiency of existing studies.Concretely,in the static personality of user modeling,arguing that the existing text-driven personality analysis researches usually ignore the fine-grained semantic information and lack the utilization of time dimension information,this paper proposes a new personality analysis model based on multi-view text features.In the dynamic capacity of user modeling,inspired by cognitive diagnosis and knowledge tracing,this paper proposes a topic-oriented capacity tracing model in MOOC such a weak-interaction scenario.In the emotional opinion of user modeling,this paper notices the characteristics of teaching evaluation text in MOOC,and proposes a cognitive process-inspired sentiment analysis model for text with continuous semantic units.For the above three types of user modeling,the proposed methods are verified by constructing real-world datasets.Secondly,for the task on knowledge concept graph construction,this paper puts forward solutions on large-scale and small-scale corpus respectively.Concretely,from the view of general knowledge graph construction based on large-scale corpus,especially for the joint extraction of entity and relation,this paper reveals the challenging issues,that is,information loss,error propagation,and ignoring the relationship between entity and relation.Subsequently,this paper proposes a new model from a new stereoscopic perspective,and conducts extensive experiments on multiple benchmark datasets.Further,this paper focuses on knowledge concept graph construction in MOOC with small-scale corpus.For the issues of few text corpora and low-frequency nested knowledge concept,this paper proposes a two-stage method on knowledge concept recognition,where candidate knowledge concepts are extracted first and then classified.For each knowledge concept pair,this paper extracts multi-source features from the aspects of language,position,learning records and so on,to predict the hierarchical and prerequisite relations.Through constructing real-world dataset,the performance on knowledge concept recognition and knowledge relation extraction is evaluated,leading to the solution on knowledge concept graph construction based on small-scale corpus.Lastly,for the task on learning path recommendation,this paper explores it from the perspective of integrating user modeling and knowledge concept graph.As a result,the methods fusing user modeling,knowledge concept graph,and both the two are proposed successively,to improve the accuracy and interpretability of recommended courses.On one hand,this paper proposes capacity tracing-enhanced course recommendation model.By introducing learner capacity into recommendation model,it provides interpretability from the perspective of capacity growth while improving the accuracy.On the other hand,this paper proposes knowledge concept path-enhanced sequential course recommendation model.By introducing knowledge concept path via meta-path rules,it provides interpretability from the perspective of course selection process reflected by metapath,while improving the accuracy.Finally,this paper proposes capacity-supervised &knowledge-enhanced sequential course recommendation model,which integrate capacity information into knowledge concept path-enhanced sequential course recommendation model.Through constructing real-world dataset,the performance on course recommendation is evaluated,leading to a unifed solution on learning path recommendation based on user modeling and knowledge concept graph in MOOC.