深度学习技术的发展为点击率预测任务带来了巨大变革，并且在商品推荐、计算广告等工业场景中取得了丰富成果。然而，基于深度学习的CTR模型在提升预测表现的同时需要消耗更多的计算资源，冗余结构的存在也严重限制了模型表现的进一步提升。基于此，研究者重新聚焦于特征工程与模型结构优化，利用网络结构搜索技术设计了兼具表现与性能的CTR模型。笔者整理了近两年顶级会议与arXiv上的相关工作。

 

包括以下论文：

  1. Neural input search for large scale recommendation models. (KDD 2020)
  2. AutoEmb: Automated Embedding Dimensionality Search in Streaming Recommendations. (ICDM 2021)
  3. Learnable embedding sizes for recommender systems.  (ICLR 2021)
  4. AutoDim: Field-aware Embedding Dimension Search in Recommender Systems. (WWW 2021)
  5. Autofis: Automatic feature interaction selection in factorization models for click-through rate prediction. (KDD 2020)
  6. Autofeature: Searching for feature interactions and their architectures for click-through rate prediction. (CIKM 2020)
  7. Autogroup: Automatic feature grouping for modelling explicit high-order feature interactions in ctr prediction. (SIGIR 2020)
  8. AIM: Automatic Interaction Machine for Click-Through Rate Prediction. (arXiv:2111)
  9. Towards automated neural interaction discovery for click-through rate prediction. (KDD 2020)
  10. AMER: automatic behavior modeling and interaction exploration in recommender system. (IJCAI 2021)

内容中包含的图片若涉及版权问题，请及时与我们联系删除