CAJ | 학술논문

传统item-based协同过滤算法根据项目之间的相似性来选取最近邻居.然而,现存的几种相似性度量方法都存在相应的弊端,因此只根据相似性无法找到准确可靠的最近邻.根据对两项目共同评分的用户个数,建立项目近邻等级,提出了结合项目近邻等级与相似性求取最近邻的新方法.另外,对于系统中新加入的项目,因为其上评分信息的匮乏,求得的最近邻往往是不准确的.为此,提出了聚合最近邻和"集体评分"两种改进方法.在MovieLens数据集上的实验结果表明,将上述改进应用于传统itembased协同过滤算法,推荐质量有明显的提高.
전통item-based협동과려산법근거항목지간적상사성래선취최근린거.연이,현존적궤충상사성도량방법도존재상응적폐단,인차지근거상사성무법조도준학가고적최근린.근거대량항목공동평분적용호개수,건립항목근린등급,제출료결합항목근린등급여상사성구취최근린적신방법.령외,대우계통중신가입적항목,인위기상평분신식적궤핍,구득적최근린왕왕시불준학적.위차,제출료취합최근린화"집체평분"량충개진방법.재MovieLens수거집상적실험결과표명,장상술개진응용우전통itembased협동과려산법,추천질량유명현적제고.
Conventional item-based collaborative filtering algorithm selects neighbors by the similarity between items.However,existing similarity measures suffer from their drawbacks respectively,thus cannot bring about accurate and reliable neighbors.A new method is proposed to select k-nearest neighbors by combining item similarity and neighbor grade based on the number of users that co-rate both items.In addition,for new items lacking rating information,two measures,aggregate neighborhood and common ratings,are presented.Experiments on the MovieLens data set show that the new algorithmic elements can help to increase the accuracy of recommendations.