SeMANTIC

This is an offical implementation of "Sample Efficiency Matters: Training Multimodal Conversational Recommendation Systems in a Small Data Setting"

Dataset

MMD-v3 dataset can be downloaded from google drive link

Configs

Dataset config can be setup in config/dataset_config.py

Training config can be setup in config/train_config.py

Run

Run experiments on MMD-v3:

# Preprocess MMD-v3 dataset
python preprocess_mmd-v3.py

# Start training
python train_mmd.py <model_file_name> <device>

Run experiments on MMD-v2:

# Preprocess MMD-v2 dataset
python preprocess_mmd-v2.py

# Start training
python train_mmd.py <model_file_name> <device>

Run experiments on SIMMC:

# Preprocess SIMMC dataset
python preprocess_simmc.py

# Start training
python train_simmc.py <model_file_name> <device>