CAJ | 학술논문

针对传统协同过滤推荐算法面临的稀疏性、实时性问题，提出了一种适用于朋友关系社交网络的社会化推荐算法。首先使用 Nystrm 扩展谱聚类方法根据朋友关系对用户进行聚类，然后在用户所属类中寻找最近邻并产生推荐；对用户进行聚类，改善了数据稀疏性问题，用户的聚类过程可离线完成，加快在线推荐速度，提高了系统实时性。在 Flixster 上的实验结果表明，与传统推荐算法相比，该算法在平均绝对偏差、覆盖率指标上都有较大改善，提高了推荐系统性能。
침대전통협동과려추천산법면림적희소성、실시성문제，제출료일충괄용우붕우관계사교망락적사회화추천산법。수선사용 Nystrm 확전보취류방법근거붕우관계대용호진행취류，연후재용호소속류중심조최근린병산생추천；대용호진행취류，개선료수거희소성문제，용호적취류과정가리선완성，가쾌재선추천속도，제고료계통실시성。재 Flixster 상적실험결과표명，여전통추천산법상비，해산법재평균절대편차、복개솔지표상도유교대개선，제고료추천계통성능。
In order to solve the sparsity and real-time problem of traditional recommendation algorithms,this paper proposed a new social recommendation algorithm for the social networks of friends.Firstly,it clustered users with Nystrm extension spec-tral cluster,mading use of the friend links.Then,it found the nearest neighbors of the user according to the user’s clusters and generated recommendations.On the one hand,spectral clustering algorithm could improve the sparsity problem .On the other hand,spectral clustering could performance off-line,which would accelerate the speed of online recommendation and im-prove the real-time problem.The experimental results on Flixter show that the new algorithm improves recommendation quality in MAE and coverage.