SETRec

This is the pytorch implementation of our paper

Order-agnostic Identifier for Large Language Model-based Generative Recommendation

Xinyu Lin, Haihan Shi, Wenjie Wang, Fuli Feng, Qifan Wang, See-Kiong Ng, Tat-Seng Chua

Environment

• Anaconda 3
• Python 3.8.0
• pytorch 2.0.1
• transformers 4.41.0

Usage

Data

The experimental data are in './data' folder, including Toys, Beauty, Sports, and Steam.

▶️ CF Tokenizer

A pre-trained CF tokenizer is utilized in SETRec to obtain the CF token. You can either

• Approach 1 - train your own CF tokenizer (e.g., SASRec)

• Approach 2 - directly use our provided CF tokens (e.g., SASRec_item_embed.pkl under each specific dataset folder)

Skip this step if you directly use our provided ones (Approach 2).

▶️ Semantic Tokenizer

A semantic tokenizer will be created and trained during the SETRec training. But before that, we need item semantic representation for tokenization. To obtain the semantic representation, you can either

• Approach 1 - extract manually The item semantic representation is extracted by pre-trained language model (e.g., T5 or Qwen). We provide the scripts for extracting semantic representation in "./data" folder, extract_item_semantic_rep.py.

• Approach 2 - use our provided semantic representations under each specific dataset folder.

Skip this step if you directly use our provided ones (Approach 2).

🔴 Training

First direct to the './code' folder

cd code

For T5 backend, run the command

bash scripts/train_t5.sh <dataset> <lr> <n_sem> <alpha>

For Qwen backend, run the command

bash scripts/train_qwen.sh <dataset> <lr> <n_sem> <alpha>

• The log file will be in the './log/' folder.
• The explanation of hyper-parameters can be found in './code/parse_utils.py'.

🔵 Inference

Get the results of SETRec by running inference.py:

• Infer with T5 backend

bash scripts/inference_t5.sh <dataset> <n_sem> <ckpt_path>

• Infer with Qwen backend (Please stay tuned for the inference script for Qwen. Currently you can get the evaluation results at the end of training.)

bash scripts/inference_qwen.sh <dataset> <n_sem> <ckpt_path>

⚪ Examples

1. Train SETRec (T5) on Toys dataset

cd ./code
bash scripts/train_t5.sh toys 1e-3 4 0.7

2. Inference

cd ./code
sh scripts/inference_t5.sh toys 4 <ckpt_path> 

Citation

If you find our work is useful for your research, please consider citing:

@inproceedings{lin2025order,
  title={Order-agnostic Identifier for Large Language Model-based Generative Recommendation},
  author={Lin, Xinyu and Shi, Haihan and Wang, Wenjie and Feng, Fuli and Wang, Qifan and Ng, See-Kiong and Chua, Tat-Seng},
  booktitle={SIGIR},
  year={2025}
}

License

NUS © NExT++