BUIR: Bootstrapping User and Item Representations for One-Class Collaborative Filtering

This is the author code of "Bootstrapping User and Item Representations for One-Class Collaborative Filtering" (SIGIR 2021).
We provide two types of user/item encoder: 1) BUIR-id - a single embedding layer, and 2) BUIR-nb - the LightGCN.

Overview

Our BUIR framework optimizes the representations of positively-related users and items (i.e., users and their highly preferred items) similar to each other, without explicitly using negative user-item interactions.

Run the codes

python
pytorch

BUIR using a simple encoder (with a single embedding layer)

python main.py --dataset toy-dataset --model buir-id --gpu <gpu-device-idx>

BUIR using a GCN-based encoder

python main.py --dataset toy-dataset --model buir-nb --gpu <gpu-device-idx>

Citation