Friday 10:45 AM–12:15 PM in Room #220/219 (2nd Floor)
Beyond Sentiment -- Emotion Mining with Python and machine learning
Max Tsvetovat
- Audience level:
- Experienced
Description
Learn how to extract emotional content from textual data - and how to build a sentiment analysis tool that does not suck.
Typical sentiment analysis tries to map the entire rich and varied world of human emotions into "good" vs "bad". In this tutorial, we use the characters of "Inside Out" and machine learning to build a nuanced model of human emotions -- and put it in production!
Abstract
Psychology recognizes 6 basic emotions -- joy, sadness, anger, disgust, fear and surprise. Sentiment analysis recognizes only two, joy and sadness -- and is notoriously inaccurate. In this talk, I will introduce students to using machine learning for emotion analysis of texts. We will train an "Inside Out" multidimensional classifier, and learn how to make a nuanced predictions of emotional content of movie reviews, touching on such stalwarts as "bittersweet melodrama", "disgusting comedy", and the potent mix of anger, fear and surprise in action movies.
We'll add a couple more dimensions to the classifier, judging emotional distance (immediacy), certainty, and other variables that constitute the latest findings in cognitive science of emotions.
We will use Python with SCLearn module, and TensorFlow to build the classifiers.
Sponsors
