摘要
协同过滤算法是一种经典的推荐算法,思想是依据近邻用户或者相似物品对目标进行推荐,常被应用在各类推荐系统中。但传统算法过分考虑热门物品对评分的影响,而忽略了冷门物品对用户兴趣特征度量的贡献,也未考虑用户兴趣动态变化的问题。对此,提出一种新的相似度改进算法,改进后的协同过滤算法将物品热门惩罚因子和时间数据权重进行加权计算,优化了用户相似度计算方法,形成了一种新的相似性度量模型。利用MovieLens电影推荐数据集验证改进后的算法,实验结果表明,该算法将推荐平均绝对误差(MAE)与传统算法相比降低了13.2%,推荐质量有了明显提升。
Collaborative filtering algorithm is a classic recommendation algorithm,which is based on the nearest neighbors or similar objects,and is extensively used in many personalized systems.However,the traditional collaborative filtering algorithms excessively consider the influence of popular objects on the scoring,but ignore the contribution of unpopular objects,and do not consider the dynamic change of user interests.In order to solve these problems,an improved similarity measurement algorithm is put forward,which is based on popular penalty factor and time data weight.The improved algorithm is validated on MovieLens dataset.The experimental results show that the MAE is reduced by 13.2%compared with the traditional algorithm,and the quality of recommendations has been significantly improved.
作者
刘超慧
韩传福
陈天成
孔先进
Liu Chaohui;Han Chuanfu;Chen Tiancheng;Kong Xianjin(School of Intelligent Engineering,Zhengzhou University of Aeronautics,Zhengzhou 450046,China)
出处
《信息技术与网络安全》
2020年第5期17-21,共5页
Information Technology and Network Security
基金
河南省科技攻关项目(182102210447)
河南省高校省级大学生创新创业训练计划项目(S201910485014)
郑州航院教研项目(zhjy18-50)。
关键词
协同过滤
推荐算法
惩罚因子
时间权重
个性化推荐
相似度融合
collaborative filtering
recommendation algorithm
penalty factor
time weight
personalized recommendation
similarity fusion