IDGenRec: LLM-RecSys Alignment with Textual ID Learning

Note: A potential issue has been detected in the current version of the implementation. Please wait until we fix the issue before using the code.

Overview

PyTorch implementation of the paper "IDGenRec: Towards LLM-RecSys Alignment with Textual ID Learning".

To better align Large Language Models (LLMs) with recommendation needs, we propose representing each item as a unique, concise, semantically rich, platform-agnostic textual ID using human language tokens.

ID Generation Example

Paper Link:

IDGenRec: Towards LLM-RecSys Alignment with Textual ID Learning

Requirements

See ./environment.txt.

Instructions

The current implementation supports only distributed training on multiple GPUs.

For Standard Sequential Recommendation

We provide four preprocessed datasets: Amazon_Beauty, Amazon_Sports, Amazon_Toys, and Yelp, under ./rec_datasets. The initially generated IDs (e.g., ./rec_datasets/Beauty/item_generative_index_phase_0.txt) are also provided.

To train the model:

Navigate to the command folder:
```
cd command
```
Run the training script:
```
sh train_standard.sh
```
You can change the --dataset to your desired dataset name.

For Foundational Training

Please download the fusion dataset from Google Drive Link and place it under ./rec_datasets, which mixes selected Amazon datasets. Please refer to the paper for dataset details.

To train the foundation model, run:

```
sh train_foundation.sh
```

Reference

@inproceedings{tan2024towards,
  title={IDGenRec: LLM-RecSys Alignment with Textual ID Learning},
  author={Juntao Tan and Shuyuan Xu and Wenyue Hua and Yingqiang Ge and Zelong Li and Yongfeng Zhang},
  booktitle={Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR)},
  year={2024}
}

agiresearch/IDGenRec