基于协同过滤的个性化选课模式研究与实践

Research and practice-based collaborative filtering personalized elective mode

针对协同过滤算法存在数据稀疏性问题及新用户问题,根据选课系统的具体情况及特殊性,比如,优秀学生可以按相似度高的邻居评价推荐,对于成绩较差的学生应参照优秀学生的选课情况对其推荐。对协同过滤算法进行改进,在学生对课程评价矩阵的基础上,抽取出一部分具有指导意义的信息,称作"优秀学生"评分矩阵,该矩阵由选课系统中所有评分数据过滤生产,代表了系统所有学生的评价信息,并随着系统中评价数据的变化而改变,利用"优秀学生"评分矩阵,根据学生的学院、专业、性别等属性,计算目标学生与"优秀学生"间的相似度;进而生成最近邻居集合;最后根据最近邻居对其生产高质量的推荐。改进后的算法在北京信息科技大学选课系统中进行实验,实验结果表明,改进后的算法在推荐效率及准确度上有明显的提高。

The problems of data sparsity and new user issue exist in collaborative filtering algorithms, depending on the circumstances and the special nature of elective system. For example, the best students can recommend a similar high evaluation neighbor, while students with poor performance should refer to the outstanding student enrollment circumstances of its recommendation. In this paper, collaborative filtering algorithm is improved on the basis of student course evaluation matrix. Part of the instructive information is extracted, called ＂outstanding students＂ scoring matrix that filters all ratings data produced by elective systemon behalf of the evaluation information system for all students. With the changes in the data evaluation systemand the use of ＂outstanding student＂ scoring matrix based on the student＇s college, major, gender and other attributes, the similarity between target students and ＂outstanding students＂ is calculated; thereby generating nearest neighbor set. Finally, high-quality recommendation based on the nearest neighbors is produced. The improved algorithm experiments in Beijing Information Science and Technology University elective system show that the modified algorithm has been significantly improved on recommendation efficiency and accuracy.