Deep Match to Rank Model for Personalized Click-Through Rate Prediction
Authors: Ze Lyu, Yu Dong, Chengfu Huo, Weijun Ren156-163
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct extensive experiments on both public and industrial datasets to validate the effectiveness of our model, which outperforms the state-of-art models significantly. |
| Researcher Affiliation | Industry | Ze Lyu, Yu Dong, Chengfu Huo, Weijun Ren Alibaba Group {lvze.lz, dongyu.dy, chengfu.huocf, afei}@alibaba-inc.com |
| Pseudocode | No | The paper describes model design and mathematical formulations but does not include structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | Our code1 is publicly available for reproducibility. 1https://github.com/lvze92/DMR |
| Open Datasets | Yes | Public Dataset Alimama Dataset2 contains ad display and click logs randomly sampled from Taobao in 8 days. It contains 26 million logs with 1.14 million users and 0.84 million items. The logs in the first 7 days is used for train set, and logs in the last day is used for test set. 2https://tianchi.aliyun.com/dataset/dataDetail?dataId=56 |
| Dataset Splits | No | For the public dataset: The logs in the first 7 days is used for train set, and logs in the last day is used for test set. For the industrial dataset: We use logs in the first 14 days as train set and logs in the following day as test set. No explicit mention of a validation set split. |
| Hardware Specification | No | No specific hardware details (like GPU models, CPU types, or memory) used for experiments are mentioned in the paper. |
| Software Dependencies | No | The paper does not specify software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow versions). |
| Experiment Setup | Yes | In the experiments on public dataset, we set learning rate to 0.001, batch size to 256, item embedding size to 32, max length of user behavior sequence to 50. The dimension of hidden layer in MLP are 512, 256, 128 respectively. Besides, the number of negative samples in the auxiliary match network is set to 2000 and the weight of auxiliary loss β is set to 0.1. In the experiments on industrial dataset, the hyperparameters are almost same as above except that we set item embedding size to 64. |