Deep-HOSeq -- ICDM 2020

Deep-HOSeq: Deep Higher-Order Sequence Fusion for Multimodal Sentiment Analysis --- ICDM 2020

Dependencies / Notes

Deep-HOSeq is written in python3 with some code fragments copied from DCGAN implementation from Carpedm20.

• The code is developed with Python 3.6 and TensorFlow 1.12.0 (with GPU support) on Linux
• For reasons of my convenience, data_dir is required to be data_dir = ../../data -- errors might pop-up when other directories are used.
• To train the model please execute main_DeepHOSeq.py

Dataset

Please download the dataset from CMU-MultimodalSDK https://github.com/A2Zadeh/CMU-MultimodalSDK

Future research (ideas)

• Better Fusion scheme for utilizing both common and unique latent information
• Utilize Sequence information for sentiment prediction
• Cross Data Generelaization Performance

Please contact either Sunny Verma or Wei Liu at firstname.lastname@uts.edu.au if you're interested to collaborate on this!