meelfy / openctr

A review and evaluation on CTR prediction models

OpenCTR

Click-through rate (CTR) prediction is an important task in many industrial applications such as online advertising, recommender systems, and sponsored search. OpenCTR builds an open-source library for benchmarking existing CTR prediction models.

Models

CTR prediction models currently available:

Model	Conference	Paper	Available
LR	WWW'07	Predicting Clicks: Estimating the Click-Through Rate for New Ads [Microsoft]	✔️
FM	ICDM'10	Factorization Machines	✔️
FTRL	KDD'13	Ad Click Prediction: a View from the Trenches [Google]
GBDT+LR	ADKDD'14	Practical Lessons from Predicting Clicks on Ads at Facebook [Facebook]
CCPM	CIKM'15	A Convolutional Click Prediction Model	✖️
FFM	RecSys'16	Field-aware Factorization Machines for CTR Prediction [Criteo]	✔️
DNN	RecSys'16	Deep Neural Networks for YouTube Recommendations [Google]	✔️
Wide&Deep	DLRS'16	Wide & Deep Learning for Recommender Systems [Google]	✔️
FNN	ECIR'16	Deep Learning over Multi-field Categorical Data: A Case Study on User Response Prediction [RayCloud]	✔️
PNN	ICDM'16	Product-based Neural Networks for User Response Prediction	✔️
DeepFM	IJCAI'17	DeepFM: A Factorization-Machine based Neural Network for CTR Prediction, [Huawei]	✔️
NFM	SIGIR'17	Neural Factorization Machines for Sparse Predictive Analytics	✔️
AFM	IJCAI'17	Attentional Factorization Machines: Learning the Weight of Feature Interactions via Attention Networks	✔️
DCN	ADKDD'17	Deep & Cross Network for Ad Click Predictions [Google]	✔️
FwFM	WWW'18	Field-weighted Factorization Machines for Click-Through Rate Prediction in Display Advertising [Oath, TouchPal, LinkedIn, Ablibaba]	✖️
xDeepFM	KDD'18	xDeepFM: Combining Explicit and Implicit Feature Interactions for Recommender Systems [Microsoft]	✖️

Datasets

Dataset	#Fields	#Features	#Instances	#Train	#Validation	#Test	Positive Ratio
Criteo	39		45,840,617	39,798,482		6,042,135	25.62%
Avazu	23		40,428,967	36,210,029		4,218,938	16.98%
iPinyou			14,758,717	12,237,087		2,521,630	0.0726%
Taobao
Alimama
Tencent
Bytedance