AT4CTR: Auxiliary Match Tasks for Enhancing Click-Through Rate Prediction
Authors: Qi Liu, Xuyang Hou, Defu Lian, Zhe Wang, Haoran Jin, Jia Cheng, Jun Lei
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct extensive experiments on one public dataset and one large-scale industrial recommendation dataset. The result demonstrates the effectiveness of the proposed auxiliary match tasks. AT4CTR has been deployed in the real industrial advertising system and has gained remarkable revenue. |
| Researcher Affiliation | Collaboration | Qi Liu1*, Xuyang Hou2, Defu Lian1 , Zhe Wang2, Haoran Jin1, Jia Cheng2, Jun Lei2 1 University of Science and Technology of China 2 Meituan |
| Pseudocode | No | The paper does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide concrete access to source code for the methodology described. |
| Open Datasets | Yes | Taobao Dataset (Zhu et al. 2018) is widely used in CTR research. |
| Dataset Splits | No | The paper describes a 'split standard' for the Taobao dataset by referencing another paper and a temporal split for the Industry dataset into training and testing. However, it does not explicitly state percentages, counts, or a dedicated 'validation' split for either dataset. |
| Hardware Specification | Yes | For the industrial dataset, ... using 8 80G A100 GPUs with the batch size 1500 of a single card. For the Taobao dataset, ... use one single 80 A100 for training with batch size 1024. |
| Software Dependencies | No | The paper states: 'We implement AT4CTR with Tensorflow.' However, it does not provide a specific version number for Tensorflow or any other key software component. |
| Experiment Setup | Yes | For the industrial dataset, the embedding size is 16 and the learning rate is 5e 4. ... For the Taobao dataset, we set the embedding size to be 18, the learning rate to be 1e 3, and use one single 80 A100 for training with batch size 1024. We set τ1 to be 0.07 and τ2 to be 0.1. We use Adam as the optimizer for both datasets. |