linxyhaha/equal

Implementation of paper: Equivariant Learning for Out-of-Distribution Cold-start Recommendation. (backbone model CLCRec) (MM'23)

Equivariant Learning for Out-of-Distribution Cold-start Recommendation

This is the pytorch implementation of our paper

Equivariant Learning for Out-of-Distribution Cold-start Recommendation

Environment

• Anaconda 3
• python 3.7.11
• pytorch 1.10.0
• numpy 1.21.4

Usage

Data

The experimental data are in './data' folder.

Training

python main.py --model_name=$1 --data_path=$2 --batch_size=$3 --l_r=$4 --reg_weight=$5 --num_neg=$6 --lr_lambda=$7 --num_sample=$8 --temp_value=$9 --dim_E=$10 --alpha=$11 --pos_ratio=$12 --neg_ratio=$13 --align_all=$14 --mse_weight=$15 --log_name=$16 --gpu=$17

or use run.sh

sh run.sh CLCRec micro-video 256 0.001 0.001 512 0.1 0.7 1 128 0.9 0.1 0.1 0 0.01 log 0

• The log file will be in the './code/log/' folder.
• The explanation of hyper-parameters can be found in './code/main.py'.
• The default hyper-parameter settings are detailed in './code/hyper-parameters.txt'.

Inference

Get the results of EQUAL with Implicit Alignment Module (IAM) by running inference.py:

python inference.py --backmodel=$1 --drop_obj=$2 --dropout=$3 --topN=$4 --log_name=$5 --gpu=$6

or use inference.sh

sh inference.sh CLCRec model [0,0.05,0.1,0.15,0.2] 100 log 0

The explanation of hyper-parameters can be found in './code/inference.py'. The default hyper-parameter settings are detailed in './code/hyper-parameters.txt'.

Examples

1. Train EQUAL on micro-video:

cd ./code
sh run.sh CLCRec micro-video 256 0.001 0.001 512 0.1 0.7 1 128 0.9 0.1 0.1 0 0.01 log 0

2. Inference on Amazon:

cd ./code
python inference.py --backmodel CLCRec --drop_obj model --dropout [0,0.05,0.1,0.15,0.2] --topN 50 --log_name log --gpu 0

equal's Issues

Complete dataset and key parameters

This is a very interesting piece of work that considers the cold start problem from the perspective of feature migration.

However, when I run the code, both datasets you provided do not run successfully, where it seems micro-video is missing key files such as validation_dict.npy; in addition, the Amazon dataset will run incorrectly due to inconsistencies in parameter inputs causing runtime errors in the calculations;

Would you mind double-checking that the key parameters of the uploaded code are consistent with the paper that
In other words, can you successfully run the code that has been uploaded?

Also, would you mind uploading the full micro-video dataset to facilitate community research?

Looking forward to your reply！😘

Kwai Dataset?

Thank you very much for your excellent work. Can you provide the complete Kwai dataset? If you are unable to release it due to copyright reasons, you can choose to send it to my email at [email protected]. Thank you very much.

user_embedding

请问Dataset.py中下面的嵌入如何获取，似乎是运行某个模型预训练得到的，期待你的回答。
best_user_embedding = np.load(dir_str+'/user_embedding.npy', allow_pickle=True)
best_item_embedding = np.load(dir_str+'/item_embedding.npy', allow_pickle=True)