CGSP(Cross-Domain Recommendation framework based on Graph Signal Processing)

This is code for CGSP(Cross-Domain Recommendation framework based on Graph Signal Processing).

Also this code includes 3 baselines of GSP-based methods(i.e, GF-CF, PGSP, LGCN-IDE) for unified domain.

Datasets

We have used two datasets : Amazon and Douban.

Specifically, we have used Amazon Movie and Music, Amazon Sports and Clothes, Douban Movie and Music, Douban Movie and Book.

More details, in cgsp/data/readme.md

How to Run

To run the code, you can execute the following command in your terminal:

• Parameter Configurations
• test_mode : src for inter-domain recommendation, tgt for intra-domain recommendation
• simple_model : cgsp-io, cgsp-oa, cgsp-ua
• a : hyperparameter alpha ( recommend 0.85 )

python main.py --dataset=<dataset name> --dtype=<src/tgt domain> --simple_model=<model name> --a=<alpha> --test_mode=<src/tgt>
python main.py --dataset=amazon --dtype=movie_music --simple_model=cgsp-io --a=0.85
python main.py --dataset=douban --dtype=movie_book --simple_model=cgsp-oa --a=0.85
python main.py --dataset=amazon --dtype=sport_cloth --simple_model=cgsp-ua --a=0.85

• Also, you can run GSP-based baselines(i.e, GF-CF, LGCN-IDE, PGSP) in a unified domain.
  • simple_model : gf-cf, pgsp, lgcn-ide
  • test_mode : merge-src for inter-domain recommendation, merge-tgt for intra-domain recommendation
  • For example,

python main.py --dataset=douban --dtype=movie_book --simple_model=gf-cf --test_mode=merge-tgt