摘要
在基于用户的协同过滤推荐算法中,传统的用户相似度计算方法并不能有效地同时解决用户共同评分项目数量、评分数值和项目热门度差异问题。为了解决上述问题,提出了一种新的用户相似度计算方法。首先通过融合权重余弦相似度和修正余弦相似度来缓解共同评分项目数量差异问题,其次引入两个修正因子,以缓解评分数值差异和项目热门度差异对用户相似度计算的影响,从而降低评分预测的误差。最后,使用MovieLens数据集进行实验,结果显示提出的改进算法在平均绝对误差(MeanAbsoluteError,MAE)方面优于现有的基准相似度算法。另外,提出的改进算法在预测用户评分的准确性和推荐质量两个方面具有良好的鲁棒性。
In user-based collaborative filtering recommendation algorithms,the calculation of user similarity has to face the problems such as the dif⁃ference of co-rated item number from common users,the difference of rating value and the difference of item popularity.To deal with these problems,a new method of calculating user similarity is proposed.Firstly,the difference of co-rated item number rated by common users could be relieved by the proposed method by combining the weighted cosine and modified cosine similarity.Secondly,the proposed method further exploits two correction factors to reduce the impact of numerical difference and item popularity difference on user similarity calcula⁃tion,which reduce the error of rating prediction.Finally,the experiment conducted on the MovieLens dataset shows that the proposed meth⁃od outperforms the baseline methods in terms of Mean Absolute Error(MAE)while improving the accuracy of user rating prediction and recommendation quality with excellent robustness.
作者
潘锦丰
叶东东
谭北海
余荣
PAN Jinfeng;YE Dongdong;TAN Beihai;YU Rong(School of Automation,Guangdong University of Technology,Guangzhou 510006)
出处
《现代计算机》
2021年第21期1-7,共7页
Modern Computer
基金
国家自然科学基金(No.61971148)。
关键词
协同过滤
用户相似度
修正因子
评分预测
MAE
Collaborative Filtering
User Similarity
Correction Factor
Rating Prediction
MAE