MSCRS: Multi-modal Semantic Graph Prompt Learning Framework for Conversational Recommender Systems

Framework

Requirements

python == 3.8.13

pytorch == 1.8.1

cudatoolkit == 11.1.1

transformers == 4.15.0

pyg == 2.0.1

accelerate == 0.8.0

Datasets

🌹🌹 We have supplemented the REDIAL and INSPIRED datasets with additional multimodal data. If you use the multimodal conversational recommendation dataset, please cite our paper~ ❤️

Processed dataset

The processed dataset can be found here (https://drive.google.com/drive/folders/1M2cIDFD_a-o1HCgGmN8pEeGgT21xo5oe?usp=drive_link)

Image

We have released the visual information obtained through web crawling. (The following image shows the posters and stills acquired by web crawling using item information from the INSPIRED dataset.)

Recommendation Task

Pretrain

ReDial

python rec/src/train_pre_redial.py

INSPIRED

python rec/src/train_pre_inspired.py

Train

ReDial

python rec/src/train_rec_redial.py

INSPIRED

python rec/src/train_rec_inspired.py

Conversation Task

python conv/src/train_conv.py

Acknowledgement

We thank VRICR, UNICRS and DCRS for providing the useful source code for the data preprocessing and prompt learning steps.