王留芳, 刘镇镇, 魏蓝,等.基于双因子混合加权相似度的协同过滤推荐算法[J].河南理工大学学报（自然科学版）,2020,39(6):133-138.

WANG L F, LIU Zhenzhen1, WEI Lan3, WU Zhengjiang1,et al.Collaborative filtering recommendation algorithm based on double-factorhybrid weighted similarity[J].Journal of Henan Polytechnic University(Natural Science) ,2020,39(6):133-138.

基于双因子混合加权相似度的协同过滤推荐算法

王留芳1,2, 刘镇镇1, 魏蓝3, 吴正江1

1.河南理工大学 计算机科学与技术学院，河南 焦作 454000;2.鹤壁汽车工程职业学院 电子工程系，河南 鹤壁 458030;3.厦门大学 信息学院，福建 厦门 361005

摘要:针对在数据稀疏和冷启动情况下，使用协同过滤推荐算法中传统相似度算法产生相似度不准确的问题，本文将传统相似算法中修正余弦相似度算法与基于用户属性的相似度算法加权混合，引入双因子作为权重，双因子用sigmoid函数定义，以阈值与读者借阅量的差值为变量，当读者借阅量大于（小于）阈值时，数据不稀疏（稀疏），修正余弦相似度算法权重自动增加（降低），基于用户属性相似度算法的权重自动降低（增加）。这种自动调整2种相似度算法权重的方法，既考虑了传统相似算法的优点，又避免其缺点。通过实验证明，改进后的算法提高了相似度计算的准确性，提高了推荐精度，在一定程度上解决了数据稀疏和冷启动下产生的相似度不准确问题。

关键词:推荐算法;协同过滤;阈值;双因子sigmoid函数;权重;用户属性

doi:10.16186/j.cnki.1673-9787.2020.6.19

基金项目:国家自然科学基金资助项目（11601129）

收稿日期:2020/01/11

修回日期:2020/03/22

出版日期:2020/11/15

Collaborative filtering recommendation algorithm based on double-factorhybrid weighted similarity

WANG Liufang1,2, LIU Zhenzhen1, WEI Lan3, WU Zhengjiang1

1.College of Computer Science and Technology，Henan Polytechnic University Jiaozuo  454000 ， Henan， China;2.Electrical Engineering Department，Hebi Automotive Engineering Professional College，Hebi  458030，Henan，China;3.School of Informatics，Xiamen University，Xiamen 361005， Fijian，China

Abstract:In the case of sparse data and cold start， the traditional similarity algorithm in collaborative filtering recommendation is used to generate the problem of inaccurate similarity. In this paper， the modified cosine similarity algorithm was weighted mixed with the similarity algorithm based on user attributes. Double-factor was introduced as the weight， and the double-factor was defined by sigmoid function， with the difference between the threshold value and the reader's borrowing quantity as the variable. When the reader borrowing was more than（less than） the threshold， the data was not sparse（sparse） ， the weight of the modified cosine similarity algorithm would automatically increase（decrease） ， and the weight of the algorithm based on user attribute similarity would automatically decrease（increase） . This method of automatically adjusting the weights of two simi larity algorithms not only considered the advantages of traditional similarity algorithms， but also avoided the disadvantages. The experiments showed that the improved algorithm had improved the accuracy of similarity calculation and improved the accuracy of recommendation， and it had solved the problems of data sparse and cold start to some extent.

Key words:recommendation algorithm;collaborative filtering;threshold;double-factor sigmoid function;weight;user attribute

基于双因子混合加权相似度的协同过滤推荐算法_王留芳.pdf