MetaMMF

This is our experiment codes for the paper:

Dynamic Multimodal Fusion via Meta-Learning Towards Micro-Video Recommendation

Environment settings

• Python 3.7
• Pytorch 1.7.0+cu101
• PyTorch Geometric 1.7.2
• Numpy 1.19.5
• Pandas 1.3.5

File specification

• data_load.py : loads the raw data in path ./dataset_sample, and the results are saved in path ./pro_data.
• data_triple.py : obtains the triplets for model training, and the results are saved in path ./pro_triple.
• GCN_model.py : implements the model framework of MetaMMF_GCN.
• Fusion_model.py : implements the model framework of MetaMMF.
• meta_layer.py : the regular layer of MetaMMF.
• cpd_layer.py : the simplified layer of MetaMMF.
• model_train.py : the training process of model.
• model_test.py : the testing process of model.

Usage

• Execution sequence

  The execution sequence of codes is as follows: data_load.py--->data_triple.py--->model_train.py--->model_test.py

• Execution results

  During the execution of file model_train.py, the epoch, iteration, and training loss will be printed as the training process:

  [1, 500] loss: 0.21953
  [1, 1000] loss: 0.15628
  [1, 1500] loss: 0.13521
  [1, 2000] loss: 0.12699
  ...

  File model_test.py should be executed after the training process, and the performance of MetaMMF_GCN will be printed:

  P@10: 0.0599; R@10: 0.2613; HR@10: 0.2018; NDCG@10: 0.1757

Dataset

Please check MMGCN for the datasets: Kwai, Tiktok, and Movielens.

Due to the copyright, we could only provide some toy datasets for validation. If you need the complete ones, please contact the owners of the datasets.