CAJ | 학술논문

针对电子商务系统中传统协同过滤推荐算法面临的稀疏性、准确性、实时性等问题，提出了一种基于用户谱聚类的协同过滤推荐算法。首先利用非负矩阵分解的方法对原始稀疏评分矩阵进行平滑处理，然后利用改进相似度的谱聚类方法将用户聚类，最后在用户所属类中寻找最近邻并产生推荐。用户谱聚类过程可离线完成，加快了在线推荐速度。在数据集MovieLens上的实验结果表明，该算法在平均绝对偏差、召回率、准确率等方面都有了较大改善，提高了推荐质量。
침대전자상무계통중전통협동과려추천산법면림적희소성、준학성、실시성등문제，제출료일충기우용호보취류적협동과려추천산법。수선이용비부구진분해적방법대원시희소평분구진진행평활처리，연후이용개진상사도적보취류방법장용호취류，최후재용호소속류중심조최근린병산생추천。용호보취류과정가리선완성，가쾌료재선추천속도。재수거집MovieLens상적실험결과표명，해산법재평균절대편차、소회솔、준학솔등방면도유료교대개선，제고료추천질량。
Abatract:Considering the sparsity,accuracy and the real-time problem of traditional collaborative filtering recommendation algorithms in electronic commerce system,a new collaborative filtering algorithm based on user spectral clustering is proposed. Firstly,it employs non-negative matrix factorization algorithm to fill the missing ratings. Then,it uses spectral clustering method of improved similarity to cluster users. Finally,it finds the nearest neighbors of the user according to the user's cluster and generates recommendations. Spectral clustering can be performed by off-line,which will accelerate the speed of online recommendation. The experimental results on MovieLens show that the new algorithm improves recommendation quality in MAE,recall and precision.