A quick way to get "good enough" sentiment analysis into your applications, this package uses character and word-level embeddings and LSTM networks to decide if a given text is either "positive" or "negative".
npm install dracula-sentiment --save
var dracula = require('dracula-sentiment'); var text = "xoxo cant wait"; // Output a 'positive', 'negative', or 'neutral' label console.log(text, dracula.analyze(text)); // To output a [negative, neutral, positive] float array console.log(text, dracula.score(text));
For best performance and accuracy, remove any non-ascii characters by converting them to their closest equivalents via
unidecode or something similar, and feed it sentence-sized chunks of text.
Tests aren't very extensive at present.
If you encounter any sentences where the classification is obviously wrong, open an issue and we'll work out a way to extend Dracula's training data so that it doesn't happen. Contributions to clean up the code and improve its style and performance are certainly welcome!
- 1.0.0 Original release
- 1.1.0 Adds a new function to get the underlying scores
- 1.1.1 Documentation and npm package updates