Research On Reinforcement Learning-based Method For Adaptive Learning Path Recommendation

The rapid development of information technology has brought about great changes in the field of education,and many online education platforms,such as MOOC,have been favored by a large number of learners due to their convenience,free,and flexible advantages.However,due to the provision of a fixed learning path for all learners,ignoring the individualized differences among learners,some learners have poor learning efficiency.Therefore,adaptive learning has received more and more attention,and this learning approach is aimed at automatically generating a learning path that meets the individual needs of each learner based on each learner’s learning preferences,goals,abilities,knowledge background,and other individual differences.Although many adaptive learning path recommendation models have been proposed by domestic and foreign scholars,there are still challenges,such as how to use various feature information to model learners’ implicit knowledge mastery level,how to effectively use learners’ knowledge mastery level to recommend personalized learning content,and how to improve the efficiency of learning path recommendation,etc.In light of these challenges,this thesis conducts research on adaptive learning pathway recommendation methods based on reinforcement learning.Specifically,the main work of this thesis is as follows:(1)A reinforcement-learning-based adaptive learning path recommendation model is designed.Introducing Markov Decision Process in the Adaptive Learning Path Recommendation Problem based on Learning Sessions.The knowledge tracing algorithm is used to model learners’ continuously changing knowledge levels,and the education knowledge graph is used to model the static knowledge structure.The knowledge level and knowledge structure are used together as the recommendation basis.Reinforcement learning algorithm is used to determine the learning items recommended at each time step,thus improving the efficiency of adaptive learning path recommendation.(2)In the reinforcement-learning-based adaptive learning path recommendation model,a knowledge-tracing model based on self-attention mechanism is applied to model learners’ changing knowledge levels.In order to prevent the recommended learning path from violating the logic of human cognition,the cognitive navigation algorithm is used to screen out a candidate set of learning items from a massive set of learning items,reducing the search space of the policy function in the subsequent recommendation process.In order to evaluate the impact of the entire path on learners and maximize learners’ learning outcomes,reinforcement learning algorithm is used as a recommender.In order to expand the number of reward signals and better measure the superiority and inferiority of recommended items,the change in learners’ knowledge level at each time step is used to calculate the reward function in the reinforcement learning algorithm.In order to improve the efficiency and accuracy of the algorithm,a learner simulator based on knowledge tracing is used to simulate the question-answering process,and the answered learning items and their corresponding answers are combined to form a new interaction record for tracking the knowledge level in the next time step.(3)In order to verify the effectiveness of the model designed in this thesis,extensive experiments were conducted on real-world datasets.Firstly,compared with classical baseline models,the results showed that the model designed achieved optimal performance on adaptive learning path recommendation problems.Then,experiments were conducted to examine the five main sub-modules of the model: knowledge level modeling,candidate learning item screening,recommender modeling,reward calculation,and learner simulator.The results confirmed that each sub-module had a positive impact on the model performance.Next,by analyzing the model hyperparameters,the impact of learning path length,the baseline in policy gradient,and multi-step in TD algorithm on model performance were verified,and the optimal hyperparameters were selected.Finally,through an example of adaptive path recommendation,the model was proved to be able to recommend more efficient and cognitively logical learning paths.A large number of experiments have verified the effectiveness of the method proposed in this thesis in the problem of adaptive learning path recommendation.