Uncertainty-aware Generative Recommendation

This repository contains the PyTorch implementation for the paper "Uncertainty-aware Generative Recommendation".

Data Preparation

Please follow the steps below to prepare the datasets and pre-trained components.

1. Download Raw Datasets

Download the raw datasets (e.g., Amazon18) from the Amazon Datasets repository.

2. Data Preprocessing

Filter and split the raw data into training, validation, and test sets:

bash ./data/_1/amazon18_data_process.sh

3. Text Embedding Generation

Convert textual data into vector representations:

bash ./data/_2/amazon_text2emb.sh

4. SIDs Generation

Train the RQ-VAE model:

bash ./data/_3/rqvae.sh

Generate item SIDs:

python ./data/_3/generate_indices.py

Model Training and Evaluation

1. Supervised Fine-Tuning (SFT)

bash ./train/SFT/run_train_SFT.sh

2. Reinforcement Learning (RL)

Align the model using the uncertainty-aware reinforcement learning framework:

bash ./train/RL/run_train_RL.sh

3. Evaluation

Evaluate the model performance on the test set:

bash ./eval/run_eval.sh