ConvFormer

The source code for our submission "ConvFormer: Revisiting Transformer for Sequential Recommendations".

Requirements

• Install Python(>=3.7), Pytorch(>=1.8), Tensorboard, Tensorflow, Pandas, Numpy. In particular, we use Python 3.9.13, Pytorch 1.12.1+cu116, Tensorboard 2.11.2, Tensorflow 1.14.1.
• If you plan to use GPU computation, install CUDA. We use CUDA 11.6 and CUDNN 8.0

Overview

ConvFormer consists of stacked LighTCN layers to extract the user preference representation from user behavior logs, followed by a dot-product scaler for recommendation.

Datasets

• We reuse the datasets that are provided in Google Drive and Baidu Netdisk. The downloaded dataset should be placed in the data folder.

Reproduction

We have three approaches to reproduce the results in the main paper.

• Open the ./assets_anonymous folder and check the log.log and tensorboard files in the corresponding model_data repo. For example, You can get the test performance of ConvFormer on the Yelp dataset via cat assets/conv_beauty/log.log | grep Test.

• In the main.py file, set the option do_eval=True, and load the corresponding .pt file in the assets_anonymous folder. In this way you can load the model trained in our environment, with training logs in the corresponding folder.

• Run the training pipeline from scratch by running amlt run main_results.yaml main_results, and pull the results by amlt results main_results -o ./assets/. You can also reproduce the full-sort performance with amlt run full_results.yaml full_results.

  • python src/main.py --model_name CONV --data_name Beauty --padding_mode 0 --conv_size 30 --full_sort 0 --batch_size 256
  • python src/main.py --model_name CONV --data_name Sports_and_Outdoors --padding_mode 0 --conv_size 30 --full_sort 0 --batch_size 256
  • python src/main.py --model_name CONV --data_name Toys_and_Games --padding_mode 0 --conv_size 30 --full_sort 0 --batch_size 256
  • python src/main.py --model_name CONV --data_name Yelp --padding_mode 0 --conv_size 30 --full_sort 0 --batch_size 256