FMGCE: Feature-filter and Multi-Graph Convolutional Encoders for Multimodal Recommendation

Introduction

The overall structure of the Feature-filter and Multi-Graph Convolutional Encoder (FMGCE), proposed by us is shown in this Figure.

Enviroment Requirement

• python 3.9
• lmdb 1.4.1
• matplotlib 3.8.2
• numpy 1.23.5
• pandas 2.1.3
• PyYAML 6.0.1
• scipy 1.11.4
• torch 1.12.0
• torch_scatter 2.1.2
• torchvision 0.13.0

Dataset

Download from Google Drive: Amazon Pre-Trained Datasets

Raw datasets can be accessed at Amazon product data

Training

For academic security reasons, FMGCE's code has not been officially made public yet.

cd ./src
python main.py

Acknowledgement

The structure of this code is based on MMRec. Thank for their work.