Based on the brief introduction to the user-based and item-based collaborative filtering algorithms, the problems related to the two algorithms are analyzed, and a new entropy-based recommendation algorithm is propose...
详细信息
ISBN:
(纸本)9780769530031
Based on the brief introduction to the user-based and item-based collaborative filtering algorithms, the problems related to the two algorithms are analyzed, and a new entropy-based recommendation algorithm is proposed. Aimed at the drawbacks of traditional similarity measurement methods, we put forward an improved similarity measurement method The entropy-based collaborative filtering algorithm contributes to solving the cold-start problem and discovering users' hidden interests. Using the practical data obtained from Movielens website and MAE metrics for accuracy measure, three different collaborativefiltering recommendation algorithms are compared through experiments. The results show that the entropy-based algorithm provides better recommendation quality than user-based algorithm and achieves recommendation accuracy comparable to the item-based algorithm. The experimental solution, the advantages of the entropy-based algorithm and future work tire discussed in detail.
In this paper a Learning Object Recommendation system is proposed. Learning Objects (LOs) in this context are reusable Web based resources (i.e. a web page, a video or images) that support some learning activity. The ...
详细信息
ISBN:
(纸本)9781424469208
In this paper a Learning Object Recommendation system is proposed. Learning Objects (LOs) in this context are reusable Web based resources (i.e. a web page, a video or images) that support some learning activity. The system follows a hybrid approach, combining two collaborativefiltering (CF) algorithms and a fuzzy inference system (FIS) defined by the instructor. This allows the instructor to adopt the role of facilitator, making recommendations when necessary, but allowing students to work together whenever possible. We propose that the final recommendation assigned to a LO, is the weighted average of the three models: Instructor, Profile and Correlation. Finally another FIS is used to determine the weights of these recommendations, the assignment of weights aims to compensate for some of the shortcomings of collaborative filtering algorithms. An experimental evaluation of this approach is presented.
暂无评论