面向相似App推荐的列表式多核相似性学习算法

Listwise Multi-Kernel Similarity Learning Algorithm for Similar Mobile App Recommendation

相似App推荐可以有效帮助用户发现其所感兴趣的App.与以往的相似性学习不同,相似App推荐场景主要面向的是排序问题.本文主要研究在排序场景下如何学习相似性函数.已有的工作仅关注绝对相似性或基于三元组的相似性.本文建模了列表式的相似性,并将三元组相似性与列表式相似性用统一的面向排序场景的相对相似性学习框架来描述,提出了基于列表的多核相似性学习算法SimListMKL.实验证明,该算法在真实的相似App推荐场景下性能优于已有的基于三元组相似性学习算法.

Similar App recommendation is useful for helping users to discover their interested Apps. Different from existing similarity learning algorithms, the similar App recommendation focuses on presenting a ranking list of similar Apps for each App. In this paper, we put emphasis on how to learn a similarity function in a ranking scenario. Previous studies model the relative similarity in the form of triplets. Instead of triplets, we model the ranking list as a whole in the loss function, and propose a listwise multi-kernel similarity learning method, referred as Sim List MKL. Experimental results on real world data set show that our proposed method SimListMKL outperforms the baselines approaches based on triplets.