This model is a new version based on GRACE and KGCL. It will contain different parts, consist of:
- KG
- CL
- LightGCN & GCN
- RS
Dataset is from Amazon datasets available in https://jmcauley.ucsd.edu/data/amazon_v2/index.html .
It's different parts will be add in next updates. Current predictable steps are:
- Read Data
- Embed texts(Descriptions) and initialize item nodes with them.
- Create KG from attributes such as genre and publisher
- CL for images
- CL for different created views
- RS part
Extra part(multi domain mode):
- Find best second domain
- Create the new graph and use LightGCN for initializing and then we have a new view
- combine view with first initialized embedding which was based on descriptions.