论文笔记:Top-N Recommendation with High-Dimensional Side Information via Locality Preserving Projection

一、基本信息

论文题目:《Top-N Recommendation with High-Dimensional Side Information via Locality Preserving Projection》

作者及单位:

论文笔记:Top-N Recommendation with High-Dimensional Side Information via Locality Preserving Projection_第1张图片

发表时间:SIGIR 2017

 

二、摘要

在本文中,我们利用高维侧信息来增强Top-N推荐。为了减少高维的影响,我们在推荐模型中引入了一种维数约简方法,即局部保持投影(LPP)。提出了一种联合学习模型,可以同时迭代地完成降维推荐任务。具体来说,推荐模型生成的项目相似度被用作LPP邻接图的权重,而投影则被用于偏差项目相似度的学习。采用LPP进行推荐,不仅保留了局部性,而且提高了项目的相似性。实验结果表明,该方法优于现有的方法。

 

三、主要内容与工作

1、Among the many available dimensionality reduction methods, Locality Preserving Projection (LPP) [3] has been shown to produce a low-dimensional space that well preserves locality. As recommendation quality largely depends on item similarity, LPP is a natural candidate in this setting.

2、

  • we propose a top-n recommendation method to harness high-dimensional side information. By introducing a projection matrix, high-dimensional side information is reduced into a low-dimensional space.
  • We present a joint learning model to simul-taneously perform LPP and learn item similarity.
  • We then conceive an alternative iterative optimization method to solve the model. 

3、论文笔记:Top-N Recommendation with High-Dimensional Side Information via Locality Preserving Projection_第2张图片

4、本文提出的模型

论文笔记:Top-N Recommendation with High-Dimensional Side Information via Locality Preserving Projection_第3张图片

 

四、总结

本文介绍了利用高维侧信息提高推荐绩效所遇到的问题,现有文献对这些问题的研究还不充分。我们提出了一种新的方法来解决这一挑战,即投影规则化项目相似性模型——Prism。该方法将LPP和Top-N推荐集成到一个联合学习算法中。在新的框架下,LPP不仅解决了高维性带来的问题,而且提高了项目相似性的相关性。我们进行了大量的实验,结果证明了Prism的优越性。

 

你可能感兴趣的:(baseline2,推荐系统会议论文)