This repository contains the source code of our paper Multi-task Item-attribute Graph Pre-training for Strict Cold-start Item Recommendation (submitted to RecSys 2023).
To use ColdGPT:
-
download data folder from: https://drive.google.com/drive/folders/1Vdu9N0p8bOW9B5cA3GgN3M5SolVY3Img?usp=sharing Place the downloaded data folder inside this folder.
-
run ColdGPT.py to pretrain a bipartite item-attribute graph. E.g.:
python ColdGPT.py --t1 --t3 --plm SBERT
- run evaluate.py to insert the SCS items into the pretained item-attribute graph. Extract the embeddings of the items for making racommendations. E.g.:
python evaluate.py --t1 --t3 --plm SBERT
Links to the Google drive folders containing all four preprocessed SCS datasets will be uploaded soon.