Research On Course Recommendation Model Based On Short-term Preference And Learning Behavior Enhancement

The rapid development of online education has exacerbated the overload of educational resources in Internet courses.With the exponential growth of online resources,it is difficult for learners to find suitable course resources in numerous online courses.Therefore,in order to better realize AI-enabled smart education,it is one of the current research hotspots in the field of smart education to recommend high-quality course resources that suit learners' learning styles from the vast amount of online resources.However,learners' learning interest has extremely dynamic,it would learning changes over time in a complex learning environment.And learners' learning of the course is a long-term and cyclical process.Meanwhile,learners' learning behaviors(this paper refers to learners' registration or selection behaviors for courses)are very sparse,but each learning behavior has a certain degree of strong and weak correlation,and traditional course recommendation systems cannot effectively alleviate the above problems.Based on this,this paper analyzes and improves on the existing classical model research,and proposes two course recommendation algorithm models in turn to solve the aforementioned two problems step by step.The main innovative work in this paper is as follows.(1)A Course Recommendation Model Based on Short-term Preference Reconstruction Behavior Contribution is proposed(Rec BC)to address the issue of dynamic changes in learners' interest in learning.The model by capturing learners' dynamic short-term preferences in learning behaviors using BI-LSTM,which used to improve the computation of the attention mechanism network's contribution of historical learning behaviors,and constructing new contributions of historical learning behaviors empowered by its dynamic interest,to addresses the dynamic issue of learning interest.The performance of course recommendation is also improved without changing the original model's recommendation strategy of collaborative filtering using long-term preferences.(2)A Course Recommendation Model Based on Short-term Preference and Learning Behavior Enhancement is proposed(St PBE)to address the problem of sparsity and strong and weak correlation of learning behaviors.The model aims to mitigate the above problems using augmented policy optimization with learned behaviors.Considering the influence of dynamic learning interests,the Rec BC model is first used as its base recommendation model,and then the existing reinforcement learning framework is used to classify weak learning behaviors with low similarity and low weight compare to learning preferences and strong learning behaviors with high similarity and high weight compare to learning preferences,and finally a relevance mapping network is introduced to convert weak learning behaviors to strong learning behaviors to enhance and optimize for Agent strategy input,to alleviate data sparsity and improve course recommendation model performance and generalization.To verify the validity of the research model,this paper conducts experiments on the MOOC dataset of Xuetang.com.The experimental results show that the Rec BC and St PBE models proposed in this paper are compared with the existing classical and advanced models,Rec BC increased by 1.07% and 1.51% in HR@10 and NDCG@10,and St PBE increased by 3.11% and 3.48% in HR@10 and NDCG@10,respectively.It is effectively proved that the correct treatment of learners' dynamic learning interest change and strong and weak correlation of learning behavior in this paper,which can improve the validity and generalization of course recommendation model to a certain extent.