RaFM: Rank-Aware Factorization Machines
Authors: Xiaoshuang Chen, Yin Zheng, Jiaxing Wang, Wenye Ma, Junzhou Huang
ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The proposed model achieves a better performance on real-world datasets where different features have significantly varying frequencies of occurrences. Moreover, we prove that the Ra FM model can be stored, evaluated, and trained as efficiently as one single FM... Experiments show the effectiveness of Ra FM in both public datasets and datasets of industrial applications. |
| Researcher Affiliation | Collaboration | Xiaoshuang Chen 1 Yin Zheng 2 Jiaxing Wang 3 Wenye Ma 2 Junzhou Huang 2 1Department of Electrical Engineering, Tsinghua University, Beijing, China 2Tencent AI Lab, Shenzhen, China 3Institute of Automation, Chinese Academy of Sciences, and University of Chinese Academy of Sciences, Beijing, China. |
| Pseudocode | Yes | Algorithm 1 Training the Ra FM |
| Open Source Code | Yes | The code of Ra FM is available at https://github.com/cxsmarkchan/Ra FM. |
| Open Datasets | Yes | Datasets for regression tasks are the Movie Lens 10M (ML 10M), 20M (ML 20M), and the Amazon movie review dataset3 (AMovie), respectively... Datasets for classification tasks are Frappe4, Movielens Tag (ML Tag), Avazu, and Criteo5, respectively... |
| Dataset Splits | Yes | All these datasets are randomly split into train (80%), validation (10%), and test (10%) sets. |
| Hardware Specification | No | The paper mentions 'distributed learning' for the industrial dataset but provides no specific details about the GPUs, CPUs, or other hardware used for any of the experiments. |
| Software Dependencies | No | The paper mentions using the 'FTRL algorithm' but does not specify any software names with version numbers (e.g., Python, PyTorch, TensorFlow, specific libraries with versions) that would be needed for reproducibility. |
| Experiment Setup | Yes | We adopt L2 regularizations for each model, and search the L2 coefficient from {1e 6, 5e 6, 1e 5, . . . , 1e 1} on the validation set. We search the ranks from {32, 64, 128, 256, 512} for each model... we tune ki by the following equation rather than grid search: ki = arg min k |log ni log Dk| |