SemEval-2013: Sentiment Analysis in Twitter

Sentence-Level Sentiment Analysis

Descriptions

One model to learn them all (both language model and sentiment analysis -> sentiment2vec)

Version

• se-v1.ipynb: using sum all LSTM output vectors to predict sentiment label.
• se-v2.ipynb: using sum all LSTM output vectors of sentiment word to predict sentiment label.
• se-v3.ipynb: using sentiment embedding instead of one-hot vector.

Data & Result

• https://www.cs.york.ac.uk/semeval-2013/task2.html
• 67% accuracy rate

Prerequisites

• tensorflow 1.2
• ipython notebook

Author

Binh Do