Tag - Based Learning Resource Recommendation System

In recent years, E-Learning has become the most important approach for accessing knowledge outside classroom. However the information overload makes it quiet hard for users to find the appropriate resources which meet their needs on the E-Learning platform. Recommender system can help users find their interest and recommend the satisfactory information to users via analyzing their historical behavior, so it is widely used in E-Learning platform.Tag is an important media between users and resources in recommender system, which is used as metadata to describe resources. On the one hand user’s interest can be reflected by tagging behavior; on the other hand tags can describe the properties of resources. Since tag has the characteristics of flexibility, convenient and socialization, it has been widely applied in the internet. Due to the randomness definition of user tags, unlimited and unrestricted tagging behavior, semantic ambiguity, polysemy, and tag redundancy often exist in tag data sets, these problems have seriously affected the quality of tag-based recommendation algorithm. The main work of this article includes the following aspects.(1) A tag-based collaborative filtering recommendation algorithm using dictionary is put forward. This algorithm is the full integration thought of the collaborative filtering algorithm and tagging system and applies the content-based filtering techniques to recommendation algorithm. This method tackles the problems of tag semantic ambiguity, polysemy, and redundancy with dictionary, after the procedure of cleaning up date sets by dictionary, not only the semantic of tag becomes clearer, but also makes the dimension of user’s interest matrix based on tag decreased significantly. First of all, user’s interest matrix can be modeled by tags generated from users, then find the user’s nearest neighbor by computing the similarity of user’s interest matrix, and then locate the nearest neighbors who had behaved resources and produce intermediate recommend results, at last compute the user’s interest with the intermediate recommend results by content-based recommendation and produce the final recommend result.(2) Analysis and validate work have been done by doing experiment on the Delicious data sets and data sets crawling form douban.com. The experimental results show that the algorithm brought up in this article has better accuracy rate and recall rate than the traditional one.(3) An E-learning platform namely learning resource recommendation system prototype using tags was devised and implemented. This prototype system can recommend the learning resources which meet learner’s demand by the history of user’s tagging behavior, meanwhile, using the classification of subjects to deal with the cold start problem in recommender system. The algorithm and prototype proposed in article provide strong reference significance and application value.