haaaein/p5

Recommendation as Language Processing (RLP): A Unified Pretrain, Personalized Prompt & Predict Paradigm (P5)

This repo presents implementation of the P5 large language model (LLM) for recommendation:

Paper: Recommendation as Language Processing (RLP): A Unified Pretrain, Personalized Prompt & Predict Paradigm (P5)
Paper link: https://arxiv.org/pdf/2203.13366.pdf

A reorganized and simplified repo of P5 called OpenP5 is also available on GitHub, which is an open-source library for benchmarking foundation models for recommendation under the Pre-train, Personalized Prompt and Predict Paradigm (P5):

Paper: OpenP5: Benchmarking Foundation Models for Recommendation
Paper link: https://arxiv.org/pdf/2203.13366.pdf
GitHub link: https://github.com/agiresearch/OpenP5

Another relevant repo regarding how to create item ID for recommendation foundation models is available here:

Paper: How to Index Item IDs for Recommendation Foundation Models
Paper link: https://arxiv.org/pdf/2305.06569.pdf
GitHub link: https://github.com/Wenyueh/LLM-RecSys-ID

Introduction

We present a flexible and unified Big Foundation Model for recommendation, which is the "Pretrain, Personalized Prompt, and Predict Paradigm" (P5) for recommendation. It unifies various recommendation tasks in a shared framework. In P5, all data such as user-item interactions, item metadata, and user reviews are converted to a common format -- natural language sequences. Specifically, P5 learns different tasks with the same language modeling objective during pretraining. Thus, it serves as the foundation model for downstream recommendation tasks, allows easy integration with other modalities, and enables instruction-based recommendation based on prompts. P5 advances recommender systems from shallow model to deep model to big model, and will revolutionize the technical form of recommender systems towards universal recommendation engine. With adaptive personalized prompt for different users, P5 is able to make predictions in a zero-shot or few-shot manner and largely reduces the necessity for extensive fine-tuning. On several recommendation benchmarks, we conduct experiments to show the effectiveness of P5. To help advance future research on Recommendation as Language Processing (RLP), Personalized Foundation Models (PFM), and Universal Recommendation Engine (URE), the source code, dataset, prompts, and pretrained P5 models are relased at this repository.

Requirements:

• Python 3.9.7
• PyTorch 1.10.1
• transformers 4.2.1
• tqdm
• numpy
• sentencepiece
• pyyaml

Usage

0. Clone this repo

   git clone https://github.com/jeykigung/P5.git
   

1. Download preprocessed data from this Google Drive link, then put them into the data folder. If you would like to preprocess your own data, please follow the jupyter notebooks in the preprocess folder. Raw data can be downloaded from this Google Drive link, then put them into the raw_data folder.

2. Download pretrained checkpoints into snap folder. If you would like to train your own P5 models, snap folder will also be used to store P5 checkpoints.

3. Pretrain with scripts in scripts folder, such as

   bash scripts/pretrain_P5_base_beauty.sh 4
   

   Here 4 means using 4 GPUs to conduct parallel pretraining.

4. Evaluate with example jupyter notebooks in the notebooks folder. Before testing, create a soft link of data folder to the notebooks folder by

   cd notebooks
   ln -s ../data .
   

Pretrained Checkpoints

See CHECKPOINTS.md.

You can also explore P5 in Hugging Face Hub (https://huggingface.co/makitanikaze/P5).

Citation

Please cite the following paper corresponding to the repository:

@inproceedings{geng2022recommendation,
  title={Recommendation as Language Processing (RLP): A Unified Pretrain, Personalized Prompt \& Predict Paradigm (P5)},
  author={Geng, Shijie and Liu, Shuchang and Fu, Zuohui and Ge, Yingqiang and Zhang, Yongfeng},
  booktitle={Proceedings of the Sixteenth ACM Conference on Recommender Systems},
  year={2022}
}

Acknowledgements

VL-T5, PETER, and S3-Rec