摘要
推荐的准确性(accuracy)和多样性(diversity)是推荐算法研究的二个重要指标,能够最大程度地满足用户的喜好.然而,基于准确性的推荐将导致推荐结果过于聚焦集中在某类特征上,使得多样性降低,导致用户选择的广度不足而整体效果不佳.针对推荐算法的两个指标之间的平衡以满足用户的需求,本文采用最大预测评分和最大内部相似度差异的两目标模型,选取极值点和膝点为隐式偏好,利用隐式偏好改进推荐方案搜索优化策略,提出了一种基于隐式偏好的多目标推荐算法.该算法利用切比雪夫距离在迭代过程中对偏好点动态标定,以引导个体收敛于隐式偏好区域,得到具有不同偏好的推荐方案.在Movielens和Netflix数据集上实验结果表明,与Item-based协同过滤推荐算法相比,该算法的推荐结果在确保准确率性能情况下多样性平均提升了38%和33.4%,新颖度平均提升了58.6%和125.4%,降低了多目标推荐算法的复杂度,有效解决了实际应用问题.
The accuracy and diversity of recommendation system are two important indicators for recommendation system that can best meet the preferences of users.However,accuracy-based recommendations will result in over-focused on certain types of features,reducing diversity and leading to a lack of breadth of choice for users and poor overall results.In order to address the balance between the two indicators,a multi-objective recommendation algorithm based on implicit preferences is proposed in this paper.The algorithm uses a two-objective model with maximum prediction raking and maximum intra-user similarity differences.The extreme and knee points are used as implicit preference points and dynamically calibrated by Chebyshev distances method during iteration to guide individuals to converge on implicit preference regions.Finally,recommendation solutions with different preferences are obtained for users.Experimental results on the Movielens and Netflix datasets show that,compared with the item-based collaborative filtering recommendation algorithm,the diversity of proposed method is increased by an average of 38%and 33.4%and the novelty is increased by an average of 58.6%and 125.4%while the precision is guaranteed.Meanwhile,the decision-making complexity of the multi-objective recommendation system is reduced and the practical application problem is effectively solved.
作者
陈宏
王丽萍
翁杭立
祝俊毅
郭海东
CHEN Hong;WANG Liping;WENG Hangli;ZHU Junyi;GUO Haidong(College of Education,Zhejiang University of Technology,Hangzhou 310014,China;School of Management,Zhejiang University of Technology,Hangzhou 310014,China;College of Computer Science&Technology,College of Software,Zhejiang University of Technology,Hangzhou 310014,China)
出处
《小型微型计算机系统》
CSCD
北大核心
2024年第4期830-837,共8页
Journal of Chinese Computer Systems
基金
浙江省基础公益研究项目(LGF21F020016)资助
浙江省联合基金项目(LZJWZ22E090001)资助.
关键词
推荐算法
准确性
多样性
多目标优化
隐式偏好
切比雪夫距离
recommendation algorithm
accuracy
diversity
multi-objective optimization
implicit preferences
Chebyshev distance
作者简介
陈宏,男,1984年生,硕士,讲师,CCF会员,研究方向为决策优化、教育信息化,E-mail:chzjut@zjut.edu.cn;王丽萍,女,1964年生,博士,教授,CCF会员,研究方向为进化计算、决策优化;翁杭立,男,1997年生,硕士研究生,研究方向为人工智能与模式识别;祝俊毅,男,1998年生,硕士研究生,研究方向为人工智能与决策优化;郭海东,男,1976年生,博士,讲师,研究方向为决策优化.
