JaydenZeng/EMMR

Mitigating Inconsistencies in Multimodal Sentiment Analysis under Uncertain Missing Modalities, in EMNLP2022

Python

EMMR

Tensorflow implementation for EMMR

Requirements

tensorflow-gpu==1.14.0
torch == 1.2.0
transformers==2.10.0
numpy==1.15.4
sklearn
python 3.6.13

Usage

download processed features from Google Drive or Baidu Disk
or process file by OpenSimle2.0, Bert, and Librosa. Refer to utils folder.
generate train.pkl, val.pkl and test.pkl

python Dataset.py

train (can modify settings in Settings.py)

python train.py

As for the IEMOCAP dataset, replacing the Datset.py, network.py and train.py in the iemocap folder.
valid: replace the testdata with valdata in nextBatch(traindata, valdata).
test

python train.py --train=False