Research On Teaching Resource Recommendation Based On Collaborative Filtering

In recent years,with the rapid development of the Internet and educational informatization,network teaching,which can provide learners with abundant teaching resources,has become a popular educational model in this information age.In spite of the conveniences brought by network teaching,there are still some deficiencies need to be improved urgently in network teaching system.For instance,the vast teaching resources often confuse users and make it difficult for them to find the interested resources.This further leads to the lost of some high quality resources and reduces user's learning efficiency at the same time.Therefore,better service strategies are imperative in order to improve user's searching efficiency.And that applying personalized recommendation technology into network teaching to offer targeted services to learners becomes a significant research focus.In this paper,the advantages and deficiencies of several personalized recommendation technologies were detailedly analyzed.Combined with characteristics of teaching resources,we chose collaborative filtering recommendation algorithm as teaching resources recommendation mechanism.And the algorithm was further improved to mitigate scalability,sparsity and cold-start problems which commonly occur in its recommendation process.The main research works include:(1)As for the scalability problem,considering the fact that similar users normally have similar resource attribute preferences,we used the maximum distance K-means clustering algorithm to cluster users.The algorithm selects the users with the greatest distance as the initial clustering centers,users with the same preferences are classified in the same cluster through offline clustering.The nearest neighbors of the target user are found in the clusters,so that the time-cost of neighbor query can be reduced.(2)With regard to the sparsity problem,we propose a recommendation model based on user rating and resource attribute preference.In the process of locating the nearest neighbor of the target user,the relationship between the users,resource and resource attributes,is well-considered.The rating score similarity and the preference of resource attributes are used to extract more similar users and improve the recommendation quality of the system.(3)For user cold start problem,a recommendation model based on user characteristics and information entropy is provided.According to the feature similarity,the nearest neighbors are found,and the information quantity of the users is measured by the information entropy.In this process,the average of the score data of the users who generate larger amount of information is selected as the basis of the new user?s prediction score.(4)Based on the MovieLens dataset,experiments are conducted to test the proposed improved algorithm,compared with the traditional collaborative filtering algorithm,the experimental results show that the proposed algorithm in this paper improves the accuracy of the recommendation.Finally,on the basis of the detailed analysis of the teaching resource construction platform,the improved algorithm is applied to the recommendation system of teaching resource and is further validated by the collected data.