AdaptSSR

The source code and data for our paper "AdaptSSR: Pre-training User Model with Augmentation-Adaptive Self-Supervised Ranking" in NeurIPS 2023.

Requirements

• PyTorch == 1.12.1
• pickle
• tqdm

Get Started

• Prepare Data
  • Download the Tecent Transfer Learning (TTL) dataset and put it under Data/.
  • Run the command bash prepare_data.sh under Code/. The script will unzip and preprocess the TTL dataset for experiements.
• Run Experiments
  • Code/run_pretrain.sh is the script for user model pre-training. You can modify the value of hyper-parameters to change the setting of experiments. Please refer to Code/AdaptSSR/pretrain.py for more options.
  • Code/run_finetune.sh is the script for fine-tuning the pre-trained user model on downstream tasks. You can modify the value of finetune_task to select the downstream task. Please refer to Code/finetune_classification.py for more options.