FudanCISL / CL4CTR

The source code of 《CL4CTR: A Contrastive Learning framework for CTR Prediction》

The most core code is class BasicCL4CTR in the BasicLayer.py, in which three SSL losses(L_cl, L_ali and L_uni) are computed to regularize feature representations.

CL4CTR is a model-agnostic and simple framework, which can be easily applied to existing CTR models, such as FM, DeepFM.

Datasets:

All four datasets are publicly available.

• Frappe
• ML-tag
• ML-1m
• SafeDriver