Research On Learning Resource Recommendation Algorithm Based On Deep Reinforcement Learning

With the continuous development of the Internet in today’s society,the number of learning resources is increasing exponentially,and the types are also diverse,such as learning resources in digital libraries,learning resources in online courses,and so on.In the face of so many learning resources,it becomes more and more difficult for students to find the most suitable learning resources,so how to accurately recommend the most suitable learning resources to students is a very necessary problem.A recommendation system is a system that recommends information(such as products,books,courses,etc.)that users are interested in to users according to their needs and interests.Traditional recommendation methods have many limitations in the process of recommendation: the recommendation process is static,ignoring the dynamic changes of user preferences;the goal of most recommendation systems is to maximize immediate benefits to get the final recommendation results,ignoring the possibility of long-term benefits;traditional The recommender system of the system tends to recommend popular items,and lacks the ability to explore items that have not been interacted with by humans or niche items.Deep reinforcement learning can further optimize the recommendation system model.The idea of deep reinforcement learning is similar to that of human learning.It is to evaluate an action by trying an action in a certain state and obtaining the reward value of the action in the subsequent results.Good or bad,this reward value is called the reward value.Applying deep reinforcement learning to sequence recommendation can effectively maximize long-term benefits,better explore results,and update dynamic changes in user preferences online.The introduction of deep reinforcement learning into recommender systems solves many problems of traditional recommender systems.Based on the existing deep reinforcement learning methods,this thesis focuses on how to introduce deep reinforcement learning into the recommendation system,and achieve better optimization of the results of the recommendation system by introducing deep reinforcement learning.This thesis firstly proposes a clustering-based hierarchical reinforcement learning model for personalized book recommendation in digital libraries.Then,Q-learning was introduced into the reinforcement learning model based on clustering,and a model to strengthen the recommender system by means of Q-learning was proposed.The main contributions of this thesis are as follows:(1)A clustering-based reinforcement learning model is proposed,and the model is used for personalized book recommendation in digital libraries.The model uses clustering and multi-layer reinforcement learning.Clustering can classify the books in the digital library and solve the problem of sparse book data in the digital library.Multi-layer reinforcement learning can find the noise data in the borrowing sequence and delete it,so that the model can more accurately predict the user’s preference and recommend corresponding books.(2)A multi-layer reinforcement learning model based on Q-learning is proposed,and the model is used to optimize the clustering-based hierarchical reinforcement learning model.The model combines Q-learning with a cluster-based hierarchical reinforcement learning model,adds Q-learning to the hierarchical reinforcement learning module,and calculates the reward value of the low-level tasks in the hierarchical reinforcement learning with the Q value in the Q-learning,and obtains More accurate sequences in noisy data.