The build-sentiment-classifier.ipynb Jupyter Notebook builds and exports a serialized Twitter sentiment classifier using Python with scikit-learn. The classifier is based on the approach of Go et al using the Sentiment140 data. The data can be downloaded from the Sentiment140 website.
The classifier has an accuracy of 84% on the test dataset consisting of several hundred annotated tweets. The training set consists of 1.6 million tweets automatically labeled by assuming that any tweet with positive emoticons, like :), were positive, and tweets with negative emoticons, like :(, were negative. This technique is called distant supervision using emoticons as noisy labels.
Chris Rawles