Neural-Sentiment-Analyzer-for-Modern-Hebrew

A neural sentiment analyser for Modern Hebrew. This project was written in PyTorch and is based on the paper: Representations and Architectures in Neural Sentiment Analysis for Morphologically Rich Languages: A Case Study from Modern Hebrew.

Results

Accuracy results (percentage of correct label predictions) for all architecture and representation choices on the test set; for the string-based vocabulary, and the char-based vocabulary.

String-based Vocabulary

Architecture	Linear	MLP	CNN	LSTM	BiLSTM
Token-Based	64.08	79.31	86.92	84.42	85.96
Morpheme-Based	62.38	78.19	85.73	82.19	85.88

Char-based Vocabulary

Architecture	Linear	MLP	CNN	LSTM	BiLSTM
Token-Based	66.15	75.42	84.77	74.42	79.35
Morpheme-Based	68.42	73.5	82.42	74.65	78.08