深度学习在推荐系统中的应用

Ensemble learning: 混合模型学习, ref1

Deep&Wide
DeepFM
XDeepFM
AiBox
DCN: Deep Cross Net, 2019 google
CF models:
- Based on Neural Collaborative Filtering (NCF) framework:
  - NeuMF: Neural Matrix Factorization (He et al, WWW'17)
  - ConvNCF: Outer Product-based NCF (He et al, IJCAI'18)
- Based on Translation framework:
  - TransRec: Translation-based Recommendation (He et al, Recsys'17)
  - LRML: Latent Relational Metric Learning (Tay et al, WWW'18)
Feature-based models:
- Based on Multi-Layer Perceptron:
  - Wide\&Deep (Cheng et al, DLRS'16),
  - Deep Crossing (Shan et al, KDD'16)
- Based on Factorization Machines (FM):
  - Neural FM (He and Chua, SIGIR'17),
  - Attentional FM (Xiao et al, IJCAI'17),
  - DeepFM (Guo et al, IJCAl'17)

2016 wide &deep

two parts: - wide component: a linear model, $y=\mathbf{w}^{T} \mathbf{x}+b$ - deep component: feed-forward neural network, $a^{(l+1)}=f\left(W^{(l)} a^{(l)}+b^{(l)}\right)$

Area Under Receiver Operator Characteristic Curve (AUC)

2016 deepFM

replace linear model with fm model