摘要
[目的/意义]旨在为提高协同过滤算法推荐质量提供参考。[方法/过程]针对协同过滤算法存在的数据稀疏性、冷启动等问题,提出基于用户多属性加权和兴趣相似度的协同过滤算法。该算法通过征集专家意见,借助层次分析法得出用户在年龄、性别、职业和邮编相似度的最优权重,计算多维属性相似度;将皮尔逊相似度作为约束系数,融入Weight Slope one算法,提取用户隐式标签,计算用户兴趣相似度;动态调整多维属性相似度和兴趣相似度,从而选取最优参数。[结果/结论]该算法有效降低了推荐系统的MAE值,在近邻数目较少和数据较稀疏情况下仍具优越性,提高了推荐质量。
[Purpose/significance]The paper is to provide references for improving the recommendation quality of collaborative filtering algorithm. [Method/process]Aiming at the existing problems such as data sparsity and cold start in collaborative filtering algorithm, the paper proposes a collaborative filtering algorithm based on users’ multi-attribute weighting and interests similarity. The algorithm uses Analytic Hierarchy Process method to gain the optimal weight of users’ ages, genders, occupations and zip codes by soliciting opinions from the experts. It also introduces the Pearson similarity coefficient as restraining coefficient into the Weight Slope one algorithm, and extracts users’ implicit labels to calculate users’ interest similarity. And it adjusts the similarities of multi-dimensional attributes and interests dynamically to choose the optimum parameters. [Result/conclusion]The algorithm reduces MAE value of recommendation system effectively, and still has superiority in the case of less neighbors and data sparsity, which improves the recommendation quality.
作者
罗海媛
章牧
Luo Haiyuan;Zhang Mu(School of Information Management, SunYat-Sen University, Guangzhou Guangdong 510006;Shenzhen Tourism College, Jinan University, Shenzhen Guangdong 518053)
出处
《情报探索》
2018年第5期1-7,共7页
Information Research
基金
国家社科基金重点项目"基于本体映射的非物质文化遗产本体词汇库研究"(项目编号:16AZD055)成果
关键词
协同过滤
冷启动
稀疏性
用户多属性加权
隐形标签
collaborative filtering
cold start
sparsity
users’ multi-attribute weighting
implicit label
