Research And Application On Personalized Recommendation Of Digital Learning Resources

After entering the Internet Web2.0 era, as the digital learning resource quantity explosive growth, its content is constantly enriched, in various ways entering into the Internet. While the learner cannot find the resources to meet their own needs from massive learning resources, the demand for personalized recommendation system is growing. Personalized recommendation is based on the preferences and interests of learners and learning behavior patterns, recommends interesting learning resources to learners.This paper proposes a new access to collecting learning behavior data, after analyzing the similarities and differences between different data collection methods, taking into account the actual application requirements. It uses a proxy server to embed scripts in the page, triggers the corresponding page events to get learners’ behavioral data in the learning process, saves the data in the database, providing data support for learners’ behavior analysis.This paper proposes a method of automatic extracting semantic of the page, introduces page semantic features in interest in learning models, taking into account that it is too single to extract learners’ behavior by only implicit learning behavior. The semantic library is generated by manual annotating a learning website,For the page that does not exist in the semantic library, on the basis of the existing semantic library, computing the most similar pages, we automatically assign it the semantic features of most similar pages.This paper realizes the learning interest model by the combination of implicit learning behavior and page semantic features. It uses the average residence time and mouse clicks to express learners’ interest preferences, and on this basis predicts the page score by means of multivariate linear regression model, combines with pages’ semantic features to calculate learners’interest preference matrix; It uses an improved K-means clustering algorithm to optimize the initial sub-cluster center selection; After classifying learners, it uses user-based collaborative filtering recommendation to recommend learning resources to learners.This paper also describes the overall design structure of the personalized recommendation application system, detailed design of the system module, including data acquisition module, learning behavior analysis and design modules, and personalized recommendation module.Finally, this paper builds personalized recommendation application system in the campus, collecting two classes’ implicit learning behavior data from different majors. Through experiments it demonstrates the feasibility of using multiple linear regression model to predict the page score, indicates the page mean residence time and mouse clicks to express interest learner preferences are feasible; we do a comparison test to determine the k value of the improved K-means clustering algorithm, the results show that the accuracy rate of the improved K-means clustering algorithm compared to the traditional K-means clustering algorithm increases slightly; For the success rate of the recommendation results we also do a comparison test, the results show the success rate will increases with the increasing number of pages within a certain range, and has a certain degree of improvement of the success rate, compared to the traditional K-means clustering algorithm.