| Online adaptive learning is a personalized education method that selects appropriate learning resources and learning methods for learners according to their cognitive state,learning ability,so that they can efficiently accomplish their learning goals and obtain the best learning efficiency.In online adaptive learning,a key fundamental issue is how to distinguish the individual differences of different learners in terms of cognitive state and learning ability.Although the contemporary methods based on educational theories have achieved certain results,the research on cognitive modeling for online adaptive learning still faces challenges such as inconsistency between the cognitive state obtained from modeling and external performance,difficulty in modeling the spatio-temporal evolution pattern of cognitive state,and unsound ways of applying cognitive state.To this end,this dissertation systematically conducts an exploratory research on the cognitive state modeling for the students in online adaptive learning.This dissertation analyzes and discusses how to model the implicit and dynamic cognitive states of learners in adaptive learning,focusing on the consistency of states and performance and the spatio-temporal evolution of cognitive states in the learning process,and finally discusses the application of cognitive states in the adaptive learning path recommendation problem.Specifically,we first focus on the consistency between the modeled cognitive states and performance,and conduct a study on the optimization of consistency in the diagnostic assessment of cognitive states.After that,we discuss the spatio-temporal evolution law of cognitive states,analyze it in two dimensions of time and space,and conduct a study on modeling the spatio-temporal influence in the dynamic cognitive state evolution tracking process.Finally,we focus on the cognitive states in adaptive learning Finally,the application of adaptive learning path recommendation based on cognitive state modeling is discussed,taking learning path recommendation as the entry point and addressing the current problem of unsound ways of cognitive state application in speculative measurement.We discuss the application of learning path recommendation based on cognitive state modeling.The research work in this dissertation is based on the problems and the data come from actual application scenarios.The main work and contributions of this dissertation can be summarized as follows:First,this dissertation investigates the optimization of consistency in diagnostic assessment of cognitive states.Starting from the assumption of monotonicity,which is used in educational theory to ensure the consistency of modeling state and external performance,we propose an item response ranking method for cognitive diagnosis to address the problem that the existing traditional diagnostic methods only consider monotonicity one-sidedly in the interaction function,which cannot guarantee consistency well.Pairwise level learning is introduced into cognitive diagnosis,and the biased order between item response responses is used in the optimization process to further optimize monotonicity and achieve the goal of enhancing consistency and improving modeling effects.Among them,in order to guarantee the consistency of the ability dimension while dealing with the unobserved responses in the response matrix,a two-branch pairwise level sampling strategy is proposed to construct training sample pairs.At the same time,the optimization objective function is redesigned to incorporate the monotonicity assumption into the optimization process of the cognitive diagnostic model.After that,the efficiency of modeling is discussed with an online incremental streaming data scenario as the entry point.In this dissertation,an incremental cognitive diagnosis framework is designed.A deep trait network is proposed to improve online data processing efficiency by obtaining trait parameters with inductive learning methods.In addition,a turning point analysis is proposed to determine the threshold value of model update to reduce the update frequency,and a momentum update algorithm is used to reduce the update time during the update process.At the same time,a stability penalty term is added to the loss function to ensure the stability of the idiosyncratic parameters.The proposed method is general and can be applied to most diagnostic models.Experiments on real data from multiple perspectives demonstrate the effectiveness of the proposed method.Secondly,the spatio-temporal influence modeling study in the dynamic cognitive state evolution tracking process is carried out to analyze the spatio-temporal evolution pattern of cognitive states.On one hand,in the temporal dimension,considering the existence of knowledge acquisition and knowledge forgetting in the learning process of learners,it is modeled by a recurrent neural network to portray the evolution pattern of local cognitive states corresponding to the knowledge learned before the single.On the other hand,in the spatial dimension,based on the important conclusion that learning on different knowledge affects each other in education theory,the focus is on how to use the knowledge structure to portray the propagation process of learning influence among knowledge in the learning process.In this dissertation,two new quantitative approaches are proposed to model the propagation process of learning influence on knowledge structures with multiple relationships.For undirected relations such as similarity relations,the synchronization propagation method is adopted,where the influence propagates bidirectionally between neighbor concepts.For directed relations such as prerequisite relations,the partial propagation method is applied,where the influence can only unidirectionally propagate from a predecessor to a successor.Meanwhile,we employ the gated functions to update the states of concepts temporally and spatially.Experiments are conducted on a large number of student practice records generated in real-world scenarios,demonstrating the effectiveness and interpretability of the method.Finally,this dissertation investigates the cognitive modeling-based adaptive learning path recommendation.The cognitive state represents the learner’s personality,and the general cognitive learning patterns of human are recorded in the knowledge structure.This dissertation proposes a Cognitive Structure Enhanced Framework for Adaptive Learning(CSEAL).By defining path recommendation as a Markov Decision Process(MDP)and applying an actor-critic algorithm,our framework can sequentially identify the right learning items to different learners.Specifically,we first utilize a recurrent neural network to trace the evolving knowledge levels of learners at each learning step.Then,we design a navigation algorithm on knowledge graph to ensure the logicality of learning paths,which can also reduce the search space in the decision process.Finally,the actor-critic algorithm is used to determine what to learn next,whose parameters are dynamically updated along the learning path.Extensive experiments with real-world data demonstrate both the effectiveness and the explanatory power of the proposed method in comparison with alternative approaches. |
