一种改进的兴趣相似度个性化推荐算法  

An Improved Personalized Recommendation Algorithm Based on Interest Similarity

作　　者：李浩 梁京章[1] 潘莹 LI Hao;LIANG Jing-zhang;PAN Ying(School of Electrical Engineering,Guangxi University,Nanning 530004,China;Information Network Center,Guangxi University,Nanning 530004,China)

机构地区：[1]广西大学电气工程学院,广西南宁530004 [2]广西大学信息网络中心,广西南宁530004

出　　处：《计算机技术与发展》2022年第12期1-6,共6页Computer Technology and Development

基　　金：赛尔网络下一代互联网技术创新项目(NGII20190108)。

摘　　要：传统的协同过滤推荐算法在进行相似度计算时主要考虑用户对物品的评分,通过评分获取用户之间的相似度,缺少对用户兴趣相似度的考虑,同时在进行相似度计算时未考虑用户自身属性的影响,其相似度计算存在一定的失真性。针对这一问题,提出一种改进的兴趣相似度个性化推荐算法,根据不同的用户对物品的兴趣会因用户的自身属性不同而存在差别,设计一种改进的兴趣相似度计算方法,在进行兴趣相似度计算时引入用户的自身属性因素,如年龄、性别等属性因素;根据用户对物品的兴趣会受到物品的热门程度的影响,提出物品热点影响率与物品属性满意度的概念,并根据物品的热点影响率与物品属性满意度在计算相似度时赋予物品不同的权重关系;根据用户的兴趣会随着时间的变化而发生改变,将时间因素加入到推荐过程中,最终通过融合时间因子的影响做出最终的评分预测。在MovieLens数据集上的实验结果表明,该算法的平均绝对误差(MAE)和均方根误差(RMSE)值更低,推荐效果更优。The traditional collaborative filtering recommendation algorithm mainly considers the users'rating of the item and obtains the similarity between users through the rating.It lacks the consideration of the users'interest similarity,and does not consider the influence of the users'own attributes in the similarity calculation,so its similarity calculation has certain inauthenticity.To address this problem,we propose an improved interest similarity personalized recommendation algorithm,and design an improved interest similarity calculation method based on the fact that different users'interest in items may differ according to their own attributes,and introduce users'own attributes such as age and gender when calculating interest similarity.Based on the fact that users'interest in items is affected by the popularity of items,we propose the concepts of item hotspot influence rate and item attribute satisfaction,and assign different weights to items when calculating similarity based on their hotspot influence rate and item attribute satisfaction.Based on the fact that users'interest changes with time,we add the time factor to the recommendation process,and finally make the final rating prediction by integrating the influence of the time factor.Experimental results on MovieLens dataset show that the average absolute error(MAE)and root mean square error(RMSE)of the algorithm are lower,and the recommendation effect is better.

关 键 词：推荐算法 协同过滤 兴趣相似度 物品热点影响率 物品属性满意度 时间因子 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]