It's a NLP research project for Sentiment Analysis task for positive, negative, neutral classes. The purpose of this work is to learn how to build Deep Learning models and compare different combinations and configurations of neural network layers. Except of Deep Learning layers also words features were used (based on BingLiu, MPQA, NRC and Sentiment 140 sentiment lexicons).
-
SemEval2017 Task 4: Sentiment Analysis in Twitter dataset (Base)
-
The Yahoo News Annotated Comments Corpus (YNACC) (for Tranfer Learning experiments)
List of main layers which used in models during experiments:
-
Keras embeddings
-
LSTM
-
Datastories embeddings
-
Bidirectional LSTM
-
Attention
-
ELMo
-
Convolutional layers