• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文会议>AMA Winter Academic Conference >Making Product Recommendations Based on Latent Topics: An Analysis of Online Purchase Data with Topic Models
【24h】

Making Product Recommendations Based on Latent Topics: An Analysis of Online Purchase Data with Topic Models

机译：根据潜在主题制作产品建议：使用主题模型进行在线购买数据的分析

获取原文
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

How do different data preparation variants of online purchase data influence the performance of topic models as rec-ommender models? For the empirical analysis purchase data from an online retailer for animal health products is used.The assortment of the firm ranges from special animal food and veterinary medicine to useful animal accessories for pets such as cats and dogs as well as larger animals like horses.The complete order history from January 2012 until January 2016 is obtained.As recommender models the two topic models Author Topic Model(ATM)and Sticky Author Topic Model(Sticky ATM)are selected.Both models include authorship information,i.e.customers'entire purchase histories instead of single orders are used.The Sticky ATM additionally takes into account customers'shopping behavior.In particular it considers that customers often stick with one category over a sequence of multiple items before they switch to another product category.Both topic models are compared to the benchmark models Unigram,Bigram and Collaborative Filtering.Unigram is the simplest method and calculates marginal probabilities across all ordered items.Bigram is the current recommender method used by the online retailer considered for analysis.Collaborative Filtering is the most frequently applied method in practice.All recommender models are tested with six different variants of data preparation.
机译：在线购买数据的不同数据准备变体如何影响主题模型的性能作为Rec-Ommender模型？对于实证分析，使用来自在线零售商的动物健康产品的购买数据。各种各样的公司范围从特殊的动物食品和兽医药物到有用的动物配件，如猫狗等宠物以及像马这样的较大的动物。从2012年1月到2016年1月的完整订单历史.As推荐模型选定了两个主题模型作者主题模型（ATM）和粘性作者主题模型（粘性ATM）。诸如IneCustryers'Enentire购买历史，代替使用单个订单。粘性ATM另外考虑到客户的商场行为。特别是它认为，客户常常在切换到另一个产品类别之前在一系列多个项目上粘在一系列中。与之相比基准模型Unigram，Bigram和协作筛选.unigram是最简单的方法，并计算所有的边缘概率订购的项目.bigram是在线零售商使用的当前推荐方法，用于分析.Collaborative过滤是在实践中最常用的方法。使用六种不同的数据准备变体进行测试。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱：kefu@zhangqiaokeyan.com

京公网安备：11010802029741号 ICP备案号：京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有

  • 客服微信

  • 服务号