Polarity Classification Demonstration

For best results, manual input before running Dracula.

Dracula is a deep, LSTM-based sequence tagger which operates on character embeddings. It can (so far) attempt to do binary polarity classification and part-of-speech tagging. Using characters means Dracula's models are small enough to run anywhere, including in the browser.

The Sentiment140 training data for this demonstration is a set of tweets stripped of emoticons. If the result comes back positive, then the addition of an emoticon like :D or :) shouldn't change the overall sentiment of tweet.

Tokenizer implementation by Myle Ott.

Important disclaimer: I'm not responsible for any tweets shown here, the contents of any external links or images or the classifications generated automatically by this network.

Now also available as a Node.js module »

Questions? Feedback? Comments? Send a tweet or email richard<at>sentimentron.co.uk

Read more about how Dracula works »

Home | Github