摘要
许多推荐技术(如协同过滤)存在以下不足,降低了用户的体验满意度和忠诚度:1)忽略了“用户兴趣和商品属性会随时间而改变”这一事实;2)过度追求预测准确性而牺牲了推荐多样性和新颖性.为此,提出一种能动态适应上述变化,同时优化推荐准确度、多样度和新颖度的互动式推荐系统.主要步骤:1)采用理想点法构造多目标优化函数;2)收集用户反馈信息,及时地更新推荐策略;3)基于多臂赌博机构建互动式推荐框架.实验表明,经过与用户不断地互动推荐,该系统的平均列表准确度、多样度和新颖度都在逐步提升.
The existing recommender systems still face challenges below,resulting in less than satisfactory user experiences.They have overlooked the fact that user preference and item attribute change over time.Moreover,they provide improvement in accuracy usually at the expense of diversity and novelty.In this direction,we propose multiple objective interactive recommender systems which can better balance the conflicts in diversity,novelty and accuracy metrics and adapt to changes of user preference and item attribute.The models rely on three main components:multi-objective optimization functions built by the methods of ideal points,dynamic prioritization schemes for weighting quality metrics and recommendation technologies modeled by the multi-armed bandit algorithm.The experimental results show that the proposed algorithms provide the capability to respond to a change in user requirements in real time,and recommend lists of personalized items that are accurate,diverse and novel.
作者
何炜俊
艾丹祥
HE Wei-jun;AI Dan-xiang(School of Management,Guangdong University of Technology,Guangzhou 520520,China)
出处
《小型微型计算机系统》
CSCD
北大核心
2021年第6期1192-1198,共7页
Journal of Chinese Computer Systems
基金
国家自然科学基金项目(71740024)资助.
关键词
推荐系统
多目标规划
多臂赌博机
互动式推荐
recommender systems
multiple objective decision making
multi-armed bandits
interactive
recommendations
