goldminer09 / prehash

PreHash

This is our implementation for the paper:

Shaoyun Shi, Weizhi Ma, Min Zhang, Yongfeng Zhang, Xinxing Yu, Houzhi Shan, Yiqun Liu, and Shaoping Ma. 2020. Beyond User Embedding Matrix: Learning to Hash for Modeling Large-Scale Users in Recommendation In SIGIR'20.

Please cite our paper if you use our codes. Thanks!

Author: Shaoyun Shi (shisy13 AT gmail.com)

@inproceedings{shi2020prehash,
  title={Beyond User Embedding Matrix: Learning to Hash for Modeling Large-Scale Users in Recommendation},
  author={Shaoyun Shi, Weizhi Ma, Min Zhang, Yongfeng Zhang, Xinxing Yu, Houzhi Shan, Yiqun Liu, and Shaoping Ma},
  booktitle={Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval},
  year={2020},
  page={319--328},
  organization={ACM}
}

Environments

Python 3.7.6

Packages: See in requirements.txt

pathos==0.2.5
tqdm==4.42.1
numpy==1.18.1
torch==1.1.0
pandas==1.0.1
scikit_learn==0.23.1

Datasets

The processed datasets can be downloaded from Tsinghua Cloud or Google Drive.

You should place the datasets in the ./dataset/. The tree structure of directories should look like:

.
├── dataset
│   ├── Books-1-1
│   ├── Grocery-1-1
│   ├── Pet-1-1
│   ├── RecSys2017-1-1
│   └── VideoGames-1-1
└── src
    ├── data_loaders
    ├── data_processors
    ├── datasets
    ├── models
    ├── runners
    └── utils

• Amazon Datasets: The origin dataset can be found here.

• RecSys2017 Dataset: The origin dataset can be found here.

• The codes for processing the data can be found in ./src/datasets/

Example to run the codes

• Some running commands can be found in ./command/command.py
• For example:

# PreHash enhanced BiasedMF on Grocery dataset
> cd PreHash/src/
> python main.py --model_name PreHash --dataset Grocery-1-1 --rank 1 --metrics ndcg@10,precision@1 --lr 0.001 --l2 1e-7 --train_sample_n 1 --hash_u_num 1024 --sparse_his 0 --max_his 10 --sup_his 1 --random_seed 2018 --gpu 0