«上一篇/Previous Article|本期目录/Table of Contents|下一篇/Next Article»

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:

29

期数:

2019年10期

页码:

196-200

栏目:

智能、算法、系统工程

出版日期:

2019-10-10

文章信息/Info

Title:

Research on Improved Collaborative Filtering Algorithm

文章编号:

1673-629X(2019)10-0196-05

作者:

徐志超; 单剑锋

南京邮电大学 电子与光学工程、微电子学院,江苏 南京 210046

Author(s):

XU Zhi-chao; SHAN Jian-feng

School of Electronics and Optical Engineering and Microelectronics,NJUPT,Nanjing 210046,China

关键词:

数据挖掘; 个性化推荐; 相似度修正参数; 用户特征属性向量

Keywords:

data mining; personalized recommendation; similarity correction parameter; user feature attribute vector

分类号:

TP393

DOI:

10. 3969 / j. issn. 1673-629X. 2019. 10. 038

摘要:

个性化推荐一直是互联网商品的重要特点,精准的个性化推荐一方面能够准确地定位市场,另一方面能够带来更好的用户体验。 尽管基于不同的应用场景下的推荐算法种类越来越多,但是推荐算法的智能性、精准性、稳定性还有待提高。 针对个性化的精准推荐需求,提出了一种基于用户的改进型协同过滤算法。 该算法主要解决由于不同用户存在不同的评价体系造成的评分偏差以及用户由于本身的特征属性(年龄、兴趣、性别)的不同造成的评分偏差,进而造成余弦相似度计算偏差变大的问题。 针对该问题,提出了一种融合型的余弦相似度计算方法,该方法包括一个相似度修正参数c和一个用户特征属性向量β,前者主要解决不同用户评价体系带来的偏差问题,后者是为了解决用户自身的特征属性不同产生的偏差问题。 根据协同过滤算法应用在电影评分推荐实验上的分析表明,改进型协同过滤算法大大提高了实验效率和推荐准确率。

Abstract:

Personalized recommendation has always been an important feature of Internet products. On the one hand,accurate personalized recommendations can accurately locate the market,and on the other hand,it can bring a better user experience. Although there are more and more types of recommendation algorithms based on different application scenarios,the intelligence,accuracy,and stability of the recommended algorithms still need to be improved. Aiming at the demand of personalized accurate recommendation, we propose a user-based improved collaborative filtering algorithm,which mainly solves the problem of score bias caused by different evaluation systems of different users and the user’s own characteristic attributes (age, interest, gender),thus resulting in larger calculation bias of cosine similarity. A fusion cosine similarity calculation method is proposed for this problem,which includes a similarity correction parameter αand a user feature attribute vector β. The former mainly solves the bias caused by different evaluation systems of different users. The latter is to solve deviation caused by the user’s own characteristic attributes. According to the analysis of the collaborative filtering algorithm applied to the film scoring recommendation experiment,the improved collaborative filtering algorithm greatly improves the experimental efficiency and recommendation accuracy.

常用功能

更新日期/Last Update: 2019-10-10