sedflix / EmoContext

A Shared Task on Contextual Emotion Detection in Text (done in exactly two days; one before mid-eval and one before end-eval)

EmoContext

A Shared Task on Contextual Emotion Detection in Text.

Usage

There are two different notebooks. Each of them requires a different configuration to run.

For notebooks/EmoContext_DeepMojiModels.ipynb use the following the method. Just a heads up - installing DeepMoji is not a trivial task.

docker build -t emo .
nvidia-docker run -it -v "$PWD":/app -p 8888:8888 emo

For notebooks/EmoContext_Elmo.ipynb, use this notebook on colab

Files Desprictiom

• notebooks/EmoContext_DeepMojiModels.ipynb: Contains experiments using DeepMoji
• notebooks/EmoContext_Elmo.ipynb: Contains experiments using Elmo (Tested on colab only)
• utills.py: contains a few important utills
• data/train.txt: Our training dataset
• data/devwithoutlabels.txt: Our testing dataset
• models/*py: contains some old file that are required anymore

Model Desciption

• Input: DeepMoji embedding of the give sentence.
• Augmentation techniques used: random word switching and removal
• DeepMoji with an increased vocabulary size of 3000 along with an augmented Training Dataset

Experiment Details:

Details all of ours experiments can be found in the following two documents.

• Slides(google slides)
• Report(pdf)

Results

Our best model had a f1 score of ~0.68 on the devwithoutlabels.txt.
More details can see on codalab competitions result. Our team name is chaicoffee.