DeepFM: A Factorization-Machine based Neural Network for CTR Prediction

Authors: Huifeng Guo, Ruiming TANG, Yunming Ye, Zhenguo Li, Xiuqiang He

IJCAI 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Comprehensive experiments are conducted to demonstrate the effectiveness and efficiency of Deep FM over the existing models for CTR prediction, on both benchmark data and commercial data.
Researcher Affiliation Collaboration 1Shenzhen Graduate School, Harbin Institute of Technology, China 2Noah s Ark Research Lab, Huawei, China
Pseudocode No The paper describes the model architecture and components using diagrams and mathematical equations, but it does not include explicit pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any statement or link regarding the public availability of its source code.
Open Datasets Yes Criteo Dataset: Criteo dataset 5 includes 45 million users click records. There are 13 continuous features and 26 categorical ones. We split the dataset into two parts: 90% is for training, while the rest 10% is for testing. 5http://labs.criteo.com/downloads/2014-kaggle-displayadvertising-challenge-dataset/
Dataset Splits No The paper states: 'We split the dataset into two parts: 90% is for training, while the rest 10% is for testing.' This specifies training and testing splits, but no separate validation split is explicitly defined with percentages or counts for the datasets.
Hardware Specification No The paper mentions 'tests on CPU (left) and GPU (right)' when discussing efficiency, but it does not provide specific details such as exact CPU or GPU models, or memory specifications used for the experiments.
Software Dependencies No The paper mentions optimizers (Adam, FTRL) and activation functions (relu, tanh) but does not provide specific software library names with version numbers, such as TensorFlow or PyTorch versions, that would be needed for replication.
Experiment Setup Yes To evaluate the models on Criteo dataset, we follow the parameter settings in [Qu et al., 2016] for FNN and PNN: (1) dropout: 0.5; (2) network structure: 400-400-400; (3) optimizer: Adam; (4) activation function: tanh for IPNN, relu for other deep models. To be fair, our proposed Deep FM uses the same setting. The optimizers of LR and FM are FTRL and Adam respectively, and the latent dimension of FM is 10.