摘要
针对视频协同过滤推荐算法多样性较低的问题,提出了一种基于多属性联合的多样性视频协同过滤推荐算法。根据用户与推荐系统的互动历史记录,判断用户是否满意系统的推荐项目,如果某个用户过去观看同一个主题的视频节目,并且不关心视频的作者,那么认为该用户对视频作者表现出较高的多样性,对视频节目主题表现出的多样性较低。采用信息熵与用户配置信息长度两个指标来评估项目各个属性的多样性,根据两个指标的组合将用户对每个项目属性的多样性分为4个象限,并且对用户多样性进行模糊化处理,以获得用户多样性对于4个象限的隶属度。在第一个阶段预测未评分项目的评分;在第二个阶段将所有项目重新排序,以提高推荐列表的多样性。最终,基于公开的Movielens 1M数据集进行了对比实验,实验结果证明本算法可实现接近top-N算法的准确率性能,同时具有一定的多样性增强效果。在推荐准确率与多样性平衡的应用场景下,设置合适的参数能够在损失较少推荐准确率的前提下,显著提高个体多样性、总体多样性与新颖性。
In order to improve the diversity of the collaborative filtering recommender system of videos,this paper proposed a diverse videos collaborative filtering recommender algorithm based on multi-property aggregate.According to the history of interaction between users and recommendation system,users are judged whether they are satisfied with the recommendation items of the system.If a user watches the videos on the same topic produced by different video authors,it indicates that this user shows high diversity to the video authors,and low diversity to the video subjects.Information entropy and user profile length are used to evaluate the diversity of each item’s attributes.According to the combination of the two indicators,the user’s diversity of each item’s attributes is divided into four quadrants,and the user’s diversity is fuzzified to obtain the membership degree of user’s diversity to the four quadrants.In the first phase,it predicts the rates of unrated items.In the second phase,it re-ranks all items,which improves the diversity of recommendation list.At last,experimental results based on the public Movielens 1M dataset show that,the proposed algorithm can realize the similar accuracy with top-N algorithm,at the same time,it enhances the diversity effectively.In the application scenario of balancing recommendation accuracy and diversity,setting appreciate parameters can improve the individual diversity,total diversity and freshness significantly with acceptable recommendation accuracy reduction.
作者
张艳红
张春光
周湘贞
王怡鸥
ZHANG Yan-hong;ZHANG Chun-guang;ZHOU Xiang-zhen;WANG Yi-ou(School of Computer Science and Engineering,Tianhe College of Guangdong Polytechnic Normal University,Guangzhou 510540,China;School of Computer&Communication Engineering,University of Science Technology,Beijing 100083,China;School of Computer Science and Engineering,Beihang University,Beijing,100191China;Beijing Institute of Science and Technology Information,Beijing 100044,China)
出处
《计算机科学》
CSCD
北大核心
2019年第8期78-83,共6页
Computer Science
基金
国家自然科学基金面上项目(61672077)
广东省教育厅教育科学规划教育信息技术研究项目(14JXN060)
广东省教育厅项目(2017SZ03)资助
关键词
电子商务
视频推荐系统
多样性增强
协同过滤推荐算法
重新排序算法
长尾分布
Electronic commerce
Video recommender system
Diversity enhancement
Collaborative filtering recommender algorithm
Re-ranking algorithm
Long tail distribution
作者简介
通信作者:张艳红(1978-),女,硕士,讲师,主要研究方向为大数据分析、推荐算法,E-mail:33582210@qq.com;张春光(1976-),男,博士,讲师,主要研究方向为人工智能、推荐算法、计算机网络;周湘贞(1976-),女,博士,副教授,主要研究方向为图像处理、推荐算法;王怡鸥(1990-),女,博士,助理研究员,主要研究方向为情报研究、智能信息处理。
