recsys
深度学习在推荐系统中的应用
YeeKal
•
•
"#recsys"
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)
- Based on Neural Collaborative Filtering (NCF) framework:
- 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)
- Based on Multi-Layer Perceptron:
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