Causal Embeddings for Recommendation

被引:172
作者
Bonner, Stephen [1 ,2 ]
Vasile, Flavian [1 ]
机构
[1] Criteo AI Labs, Paris, France
[2] Univ Durham, Dept Comp Sci, Durham, England
来源
12TH ACM CONFERENCE ON RECOMMENDER SYSTEMS (RECSYS) | 2018年
关键词
Recommender Systems; Causality; Embeddings; Neural Networks; Counterfactual Inference;
D O I
10.1145/3240323.3240360
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Many current applications use recommendations in order to modify the natural user behavior, such as to increase the number of sales or the time spent on a website. This results in a gap between the final recommendation objective and the classical setup where recommendation candidates are evaluated by their coherence with past user behavior, by predicting either the missing entries in the user-item matrix, or the most likely next event. To bridge this gap, we optimize a recommendation policy for the task of increasing the desired outcome versus the organic user behavior. We show this is equivalent to learning to predict recommendation outcomes under a fully random recommendation policy. To this end, we propose a new domain adaptation algorithm that learns from logged data containing outcomes from a biased recommendation policy and predicts recommendation outcomes according to random exposure. We compare our method against state-of-the-art factorization methods, in addition to new approaches of causal recommendation and show significant improvements.
引用
收藏
页码:104 / 112
页数:9
相关论文
共 31 条
  • [1] [Anonymous], 2016, P 33 INT C MACH LEAR
  • [2] [Anonymous], 2011, Proceedings of the 28th International Conference on International Conference on Machine Learning
  • [3] [Anonymous], 2016, ARXIV160603976
  • [4] [Anonymous], 1998, REINFORCEMENT LEARNI
  • [5] [Anonymous], 2007, P KDD CUP WORKSH NEW
  • [6] Berman R., 2016, Beyond the Last Touch: Attribution in Online Advertising
  • [7] Bottou L, 2013, J MACH LEARN RES, V14, P3207
  • [8] Courty Nicolas, 2016, IEEE T PATTERN ANAL
  • [9] Deep Neural Networks for YouTube Recommendations
    Covington, Paul
    Adams, Jay
    Sargin, Emre
    [J]. PROCEEDINGS OF THE 10TH ACM CONFERENCE ON RECOMMENDER SYSTEMS (RECSYS'16), 2016, : 191 - 198
  • [10] E-commerce in Your Inbox: Product Recommendations at Scale
    Grbovic, Mihajlo
    Radosavljevic, Vladan
    Djuric, Nemanja
    Bhamidipati, Narayan
    Savla, Jaikit
    Bhagwan, Varun
    Sharp, Doug
    [J]. KDD'15: PROCEEDINGS OF THE 21ST ACM SIGKDD INTERNATIONAL CONFERENCE ON KNOWLEDGE DISCOVERY AND DATA MINING, 2015, : 1809 - 1818