FinalMLP: An Enhanced Two-Stream MLP Model for CTR Prediction
Authors: Kelong Mao, Jieming Zhu, Liangcai Su, Guohao Cai, Yuru Li, Zhenhua Dong
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our evaluation results on four open benchmark datasets as well as an online A/B test in our industrial system show that Final MLP achieves better performance than many sophisticated two-stream CTR models. Our experimental results on four open benchmark datasets show that Final MLP outperforms the existing two-stream models and attains new state-of-the-art performance. Furthermore, we validate its effectiveness in industrial settings through both offline evaluation and online A/B testing, where Final MLP also shows significant performance improvement over the deployed baseline. |
| Researcher Affiliation | Collaboration | Kelong Mao1*, Jieming Zhu2*, Liangcai Su3, Guohao Cai2, Yuru Li2, Zhenhua Dong2 1Gaoling School of Artificial Intelligence, Renmin University of China 2Huawei Noah s Ark Lab 3Tsinghua University |
| Pseudocode | No | The paper does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | Our source code will be available at Mind Spore/models and Fuxi CTR/model zoo. ... To promote reproducible research, we open sourced the code and running logs of Final MLP and all the baselines used. |
| Open Datasets | Yes | We experiment with four open benchmark datasets, including Criteo, Avazu, Movie Lens, and Frappe. We reuse the preprocessed data by (Cheng, Shen, and Huang 2020) and follow the same settings on data splitting and preprocessing. Table 1 summairies the statistics of the datasets. |
| Dataset Splits | Yes | We reuse the preprocessed data by (Cheng, Shen, and Huang 2020) and follow the same settings on data splitting and preprocessing. |
| Hardware Specification | No | The paper mentions 'in parallel in GPUs' but does not provide specific hardware details such as GPU models, CPU models, or memory amounts used for the experiments. |
| Software Dependencies | No | The paper mentions 'Fuxi CTR (Zhu et al. 2021)' and 'Mind Spore' but does not specify their version numbers or other software dependencies with versions. |
| Experiment Setup | Yes | Our evaluation follows the same experimental settings with AFN (Cheng, Shen, and Huang 2020), by setting the embedding dimension to 10, batch size to 4096, and the default MLP size to [400, 400, 400]. For Dual MLP and Final MLP, we tune the two MLPs in 1 3 layers to enhance stream diversity. We set the learning rate to 1e 3 or 5e 4. We tune all the other hyper-parameters (e.g., embedding regularization and dropout rate) of all the studied models via extensive grid search (about 30 runs per model on average). |