Research On Learning Resource Recommendation Method Based On User Interest

With the rapid development of the Internet and web2.0 technology,the rapid growth of information resource on the network has caused the problem of "information overload" become more serious.It becomes very difficult for users to find information that meet their personalized needs from massive amounts of network resources such as text,video,images,and music.The recommendation system is one of the key technologies to solve the above problem.The recommendation system obtains the user's interest preferences through statistical analysis and research on the user's historical behavior data,so as to make personalized recommendation.The current application areas of recommendation systems mainly focus on e-commerce,TV movies,music websites,news information and advertising push.In other fields,there is an urgent need to recommend a system to solve problem.In the field of education,according to the "Statistical Report on the Development Status of China's Internet" and iiMedia Research survey data,with the development and popularization of the development concept of "Internet + Education" The scale of online education platforms and users has gradually increased,reaching 309 million people in 2020,and China's online education market will reach 723.06 billion yuan in 2020.How learners can quickly find content that meets their needs from the massive learning resources has become a problem that has to be faced and solved.At present,research on learning resource recommendation method is at an initial stage.and this paper propose new method for the problems of Learning Resource Recommendation,the main work of this paper are as follows:1.Aiming at the traditional learner "cold start" and data sparseness problem of traditional collaborative filtering method.A cross-domain learning resource recommendation method based on user interest transfer is proposed.Given that learners have accumulated a lot of scoring data in different fields,and the preferences of learners in different fields are similar,this method uses learner scoring information to propose a new RF-ILF method to build user interest;for the target field " "Cold start" learners,using the transformation matrix method to obtain their interest in the target field with the help of the learner 's interest in the assistance;finally,the learner 's interest in the auxiliary field is transferred to the target field to ease thetarget field data Sparsity.Experimental results on real data sets show that this method has higher accuracy,recall and F1 value than existing single-domain collaborative filtering method and cross-domain method based on scoring mode.2.The sparseness of the score data in the traditional recommendation method based on score data seriously affects the recommendation performance.In view of the learner 's comment text on the learning resource,it not only intuitively describes the reason for the item 's score and the score,but also different learners learn the same Resources have different evaluation points and narrative methods.In order to alleviate the sparseness of data and improve the accuracy of recommendation,a method of learning resource recommendation based on BERT model and comment text is proposed.Use the BERT model to process the review texts of learners and learning resources,and combine the attention mechanism to obtain learner preference vectors and learning resource feature vectors;obtain the latent hidden vectors of learners and learning resources according to the scoring matrix,and apply the above learning The two vectors of the learner and the learning resource are fused using a fusion gate;finally,the traditional prediction method of multiplying vector scores is improved to a deep neural factorization machine.The experimental results on 4 sets of data sets show that this method is superior to the traditional methods based on scoring and review text in RMSE,and the highest drop is 4.2%.