MOOC Recommendation Method Based On Improved LDA Topic Model

With the global epidemic in full swing in 2020,our government,universities and other relevant institutions actively study the situation and start to implement online teaching,and online learning has become an indispensable form of teaching in the current era.China University MOOC(hereinafter referred to as MOCC)is the representative platform of this new form of teaching.MOOC platform learning resources are extensive,covering a full range of disciplines,students,newcomers to the workplace can find their own interest in the course,its course providers to major universities,supplemented by training teams.The courses offered by universities are free of charge,and the background of the relevant universities is used for quality assurance,which is welcomed by the majority of learners.More and more teachers and experts are releasing their own courses on MOOC,which brings more course choices and also increases the information overload of the platform.To cope with this challenge,research on catechism recommendation has become a hot topic in recent years.In this thesis,by crawling 10150 course information data and 1048575 corresponding course review and rating related data of the mu course platform,through data cleaning,we get the rating and review research base data of the course.The following three aspects of research work were carried out on this basis:(1)When processing the review data,the LDA model was used to generate the "topicword" matrix,and the Word2 Vec model was introduced to convert this matrix into a "topicword vector" matrix.(2)The word vector has numerical characteristics and the topic classification has textual characteristics.The K-Prototype model is used to combine the two,completing the clustering operation of this matrix,where the topics of the reviews carry semantic information,thus calculating the course similarity matrix of the improved LDA model.(3)Using multiple linear regression,the course similarity matrices based on ratings and the improved LDA review topics are aggregated,and the test set rating matrix prepared in advance is multiplied by the course similarity matrix we have aggregated to produce a more accurate prediction score matrix,which in turn produces course recommendations.In this thesis,using MAE and RMSE as evaluation indexes,by comparing the MOOC recommendation algorithm based on course rating similarity,MOOC recommendation algorithm based on LDA topic similarity,and MOOC recommendation algorithm based on improved LDA topic model detailed in this thesis research internally,the MAE values were5.1921,4.5066,and 3.2031,and the RMSE values were 5.2722,4.6004,3.2869;meanwhile,reconstructing five datasets with Slope One algorithm and TMCF algorithm for external comparison,the improved algorithm in this thesis has the best overall performance in RMSE value,while its course prediction score performs better when the average number of course reviews increases,which proves that the proposed algorithm in this thesis can help for MOOC recommendation.