Efficient Multi-task Prompt Tuning for Recommendation

This repository contains the source code for the paper "Efficient Multi-task Prompt Tuning for Recommendation." The project aims to improve the performance of multi-task learning in recommendation systems and enable rapid generalization to new tasks by leveraging knowledge from previously learned tasks. The code related to multi-task learning is placed in the two_task branch, while the code for new task generalization is in the three_task branch.

Environment Requirements

• Python Version: 3.10
• Other Library Versions: See requirements.txt

Running Instructions

STEP 1: Prepare Dataset

1. Create a dataset directory in the root folder, and then create a Census-income directory inside it to store the Census-income dataset.
2. Download the dataset from UCI Machine Learning Repository.
3. Preprocess the data using multitaskrec/data_preprocessing/CensusIncome_process.py.

STEP 2: Multi-task Learning (Two Tasks)

1. Switch to the two_task branch:

   git checkout two_task
   

2. Run the censusincome_main.py script:

   python censusincome_main.py
   

STEP 3: New Task Generalization (Pre-train on Two Tasks and Generalize to a New Task)

1. Switch to the three_task branch:

   git checkout three_task
   

2. Run the CensusIncome_NewTask.py script:

   python CensusIncome_NewTask.py
   

Other datasets and baselines follow a similar approach.

Citation

If you find this research helpful, please cite our paper:

@article{bai2024efficient,
  title={Efficient Multi-task Prompt Tuning for Recommendation},
  author={Bai, Ting and Huang, Le and Yu, Yue and Yang, Cheng and Hou, Cheng and Zhao, Zhe and Shi, Chuan},
  journal={arXiv preprint arXiv:2408.17214},
  year={2024}
}