Research On MOOC Resource Recommendation Method Based On Interactive Feature Expression

With the rapid development of the Internet,MOOC education has become a mainstream online teaching method,gathering a large number of users and high-quality educational resources.In order to meet the personalized learning needs of users,effective course recommendation methods have become a research hotspot.On the MOOC platform,there may be multiple teaching videos corresponding to the content of a course,and the emphasis of each video may be different.How to recommend a satisfactory video to users requires us to fully model user information and explore user interests.In view of the current situation of online education recommendation,this paper conducts an in-depth study and proposes a MOOC resource recommendation method based on interactive feature expression.The main contents are as follows:（1）Proposed a video feature expression model MOOCrec based on meta-path attention mechanism.On the one hand,the relationship between learners and MOOC resources can be described by heterogeneous information network,and then the metapath can be used to express the interactive relationship between students and videos.On the other hand,attention mechanism can capture the influence of the characteristics of students,videos and meta-paths on learning interest.Specifically,MOOCrec model consists of two attention mechanisms: the first layer is the attention layer of nodes,which combines the features of nodes themselves by weighting the features of neighbors,and obtains the feature representation of entities under the meta-path by using multi-attention.The second layer is the path attention layer,which integrates the feature representations of entities learned under the guidance of different meta-paths to capture the feature representations of entities under different interests.（2）Proposed MOOC resource recommendation method based on MA-GNN interest model.This method is mainly composed of three modules: user short-term interest mining,user long-term interest mining and interest fusion module;GNN is used to capture users’ short-term interests based on the transfer diagram between items in the short-term interaction sequence.Using the attention mechanism with memory network to capture the long-term interest of users;Finally,the door mechanism similar to that in LSTM model is adopted for interest fusion.This model can fully explore user interest and improve the accuracy of recommendation.（3）In MOOCCube and MOOCdata dataset validated the recommended results based on MOOCrec model and MA-GNN model.Experimental results show that compared with other entity embedding models,MOOCrec model can fully mine semantic information and has richer feature representation effect.It is not accurate to model the common interests of users.Classifying users’ interests into long-term and short-term interests can improve the accuracy of recommendations.Furthermore,using a fusion mechanism to combine long-term and short-term interests can also improve performance.