Controlling Diversity at Inference: Guiding Diffusion Recommender Models with Targeted Category Preferences

Gwangseok Han*, Wonbin Kweon*, Minsoo Kim, Hwanjo Yu

Accepted to KDD 2025 Research Track!

Original implementation for paper

Getting Started

Python 3.10

Import conda environment

conda env create -f d3rec_env.yml

Real-world experiments

preprocess (It is already completed)

1. Move ./datasets/[dataset name]/clean_df_C20.pt to ./Real-world/datasets/[dataset name]/
2. Run cd Real-world
3. Run python preprocessing.py --dataset_name [dataset name]
4. Confirm 'Clean_C20_[6,2,2].pt' file.

Training

1. Run bash train.sh
   • Tune hyper-parameters in train.sh
   • The best hyperparameter is noted in the comments.

Inference

1. Run bash inference.sh
   • Set the hyper-parameters of the best model

Semi-synthetic experiments

• Similar to the real-world experiments