rocksat / deepmoji

CS229 course project deep learning model for analyzing sentiment, emotion, sarcasm etc

DeepMoji

CS229 course project deep learning model for analyzing sentiment, emotion, sarcasm etc

Download GLoVe Pre-trained model

foo@bar:~$ bash models/download_glove.sh

Download NLTK stop words

foo@bar:~$ python -m nltk.downloader all

Train:

• train [NB / SVM] classifier with [bow / GLoVe] word embedding

foo@bar:~$ python src/train.py -d $DATAST_PATH -t $TEST_RATIO -c $CLASSFIER_TYPE -w $WORD_EMBEDDING_TYPE -o $ARTIFACT

• train [CNN / LSTM / GRU] classifier with [GLoVe-50d / GLoVe-300d / BERT] word embedding

foo@bar:~$ python src/dnn_train.py -d $DATAST_PATH -t $TEST_RATIO -c $CLASSFIER_TYPE -w $WORD_EMBEDDING_TYPE -o $ARTIFACT

Predict:

foo@bar:~$ python src/predict.py -m models/nb.pkl -d models/word_dictionary.json -e data/emoji_map_1791.csv -s "I am happy"

Accuracy

Word Embedding	BoW + TF-IDF	GLoVe-50d	GLoVe-300d	BERT
Naive Bayes	19.530%	N/A	N/A	N/A
SVM	9.195%	16.376%	14.966%
CNN	N/A	15.168%	15.906%
LSTM	N/A	15.705%	15.570%
LSTM + Attention	N/A

To-do List

• implement BERT as word-embedding
• implement attention mechanism