数据挖掘中基于最小遗憾度的偏好感知算法

PREFERENCE PERCEPTION ALGORITHM IN DATA MINING BASED ON REGRET MINIMISATION

交互学习是数据挖掘的一种重要手段。使用交互学习作为学习系统和用户的交互模型,以向用户提供最大效用结果为目标,通过对用户反馈质量进行定量描述,考察偏好反馈,提出一种基于最小遗憾度的偏好感知算法。此外,还对偏好感知算法的期望遗憾度界限进行分析,并给出该算法的几个扩展版本。最后利用电影推荐任务及网络搜索排名数据验证了该算法的有效性。

Coactive Learning is an important means of data mining. We use coactive learning as the model of interaction between learning system and users, and propose a regret minimisation-based preference perception algorithm targeted at providing for users the maximum utility results through making the quantitative description on the quality of user feedback and studying the preference feedback. Besides, we also analyse the boundary of the expected regret of the preference perception algorithm, and give several extended versions of the algorithms. Finally, we verify the applicability of the algorithm by using movie recommendation task and ranking data of web-search.