CAJ | 학술논문

协同过滤推荐算法分为基于内存和基于模型的推荐算法，协同过滤推荐算法存在数据稀疏性、可扩展性、冷启动等问题。通过基于用户、基于项目协同过滤推荐算法以及SVD、Slope-One、KNN等基于模型协同过滤推荐算法对比分析。提出加入特征向量维度优化的SVD算法，通过降维改善数据稀疏性问题。利用Hadoop分布式平台改善推荐算法可扩展性问题。基于MovieLens数据集实验结果表明，引入基于Boolean相似性计算方法的推荐效果更优，引入数量权重和标准差权重的优化 Slope-One 算法和引入特征向量维度的优化 SVD 算法推荐效果更优。
협동과려추천산법분위기우내존화기우모형적추천산법，협동과려추천산법존재수거희소성、가확전성、랭계동등문제。통과기우용호、기우항목협동과려추천산법이급SVD、Slope-One、KNN등기우모형협동과려추천산법대비분석。제출가입특정향량유도우화적SVD산법，통과강유개선수거희소성문제。이용Hadoop분포식평태개선추천산법가확전성문제。기우MovieLens수거집실험결과표명，인입기우Boolean상사성계산방법적추천효과경우，인입수량권중화표준차권중적우화 Slope-One 산법화인입특정향량유도적우화 SVD 산법추천효과경우。
The collaborative filtering recommendation algorithm is divided into user-based and item-based recommendation algorithms. Collaborative filtering recommendation algorithm had data-sparseness and scalability and cold-start problems. This paper mainly studied the collaborative filtering recommendation algorithm based on the users or Items and SVD, Slope-One, KNN. The optimization of SVD algorithm which considers the dimension of the feature space used dimension reduction to improve data-sparseness problem. Using the Hadoop distribution platform to improve the scalability problem. Experimental result shows that the similarity computation method based on Boolean data has better result and the optimization of Slope-One and SVD algorithm have better recommendation result based on MovieLens data set.