摘要
文章针对传统推荐系统在数据稀疏性情况下用户相似性度量精度不高的问题,提出了一种多维度用户相似性度量的协同过滤算法。首先根据用户-项目打分矩阵,考察用户共同评分项目数和用户活跃度对用户相似的影响,并计算用户的相似度;然后通过修正皮尔逊用户相似性计算用户的相似性;最后通过一个权值来控制两者的重要程度,综合计算用户的相似性。研究结果表明权重系数为0.4,即修正的皮尔逊用户相似性的占的比重较大时,推荐系统的推荐质量最好;同时多维度用户相似性度量的协同过滤推荐算法在MAE、召回率RE和准确率三个方面都要优于经典的余弦相似性协同过滤算法以及皮尔逊相似性协同过滤算法。
Aimed at the problem that the accuracy of user similarity measurement is not high in the case of data sparsity in traditional recommendation system, this paper proposes a collaborative filtering algorithm for multi-dimension user similarity. Firstly, the paper relies on the user-item scoring matrix to investigate the influence of the number of collective rating items and the degree of user activity on the similarity of the users, and calculate the similarity of the users. And then the paper calculates the similarity of the user by modifying the Pearson user similarity. Finally, the similarity of the user is calculated synthetically and the importance of both is controlled by a weight. The research results show that when the weighting coefficient is 0.4, which means that the modified Pearson user similarity accounts for a large proportion of the recommendation, the recommended quality of the recom- mendation system is the best; at the same time, the collaborative filtering recommendation algorithm of multi-dimension user simi- larity measure is superior to the classical cosine similarity cooperative filtering algorithm and the Pearson similarity collaborative filtering algorithm in the aspect of MAE, recall rate and accuracy rate.
作者
王明佳
韩景倜
Wang Mingji;Han Jingti(School of Mathematics;School of Information Management and Engineering;Institute of Financial Science and Technology;Key Laboratory of Shanghai Financial Information Technology Research. Shanghai University of Finance and Economics, Shanghai 200433, China)
出处
《统计与决策》
CSSCI
北大核心
2018年第9期66-69,共4页
Statistics & Decision
基金
国家自然科学基金资助项目(71271126
61374177
61773248)
关键词
协同过滤
推荐系统
多维度
用户相似性
推荐精度
collaborative filtering
recommendation system
multi-dimension
user similarity
recommendation accuracy
