Heuristic Exercise Recommendation Model Based On Deep Reinforcement Learning

In the learning process,students'ability to learn in the "stretch zone" is always growing rapidly.Digging learners'learning behaviors and guiding learners to answer exercises in the "stretch zone"are very important for their learning process.At present,the exercise recommendation system of online education often adopts a linear organizational structure.In the absence of teacher-directed guidance,there is a low recommendation efficiency,an unstable response rate,and has a problem that is too difficult or too easy for learners to easily give up.At the same time,the personalized learning recommendation system still can't use reasonable rules and analytical models to accu-rately describe the evolution of learner learning.Since deep reinforcement learning introduces deep neural networks based on reinforcement learning,and has the ability of deep learning per-ception and decision-making for reinforcement learning,it is precisely the decision-making way in which experienced teachers can apply to multiple levels of student groups.It is used as an auxiliary problem-solving system to accurately understand the user's level,make rational thinking and make quick decisions,and guide learners to learn in the "stretch zone".This thesis takes advantage of the characteristics of deep reinforcement learning algorithms,and builds a exercise network map,combining the concept of cut sets in the learning area in psychol-ogy and in complex networks.It transforms the "stretch zone" in the process of finding learners'answers to a problem based on learner behavior.In the network diagram,the strategic problem of finding user cut sets was explored,and a heuristic exercise recommendation model based on deep reinforcement learning was constructed.The main contents are:(1)According to the relationship between the knowledge point and the title of knowledge in the exercise bank and the relationship between the knowledge points,the relationship between the exercises under the knowledge point is determined,and the exercise network diagram is con-structed;according to the user behavior data,the subgraph in the current state of the specified user is obtained in the exercise network diagram,as a learning boundary;And based on the theory of"stretch zone" in psychology,constructs a recommendation deviation metric;(2)Using the deep reinforcement learning algorithm,making use of user history record model-ing,through the action set,environmental state representation,instant reward strategy,and neural network structure design,the user's task of selecting cut set will be converted form into a deep reinforcement learning algorithm through the user's behavior.Training how to select the cut set strategy in the subgraph under the user's current state.Experiment shows that with the training of the model,the recommended deviation metrics tend to be stable,which proves the effectiveness of the model.The model based on deep reinforcement learning in heuristic coaching scenario in this thesis can be used to fit the thinking mode of empirical teachers' decision-making,extract valid implicit information in historical answer data,recommend the best exercise for learners,and can also be widely used in similar heuristic coaching scenario.