TagMe! – Building a sentiment classifier for real time tweets from scratch and crowd-sourcing it
Presented by: Abhiram
8 years ago
| 17 interested

tl;dr – TagMe is a near-real time sentiment classifier for social media content involving Twitter API, data mining and machine learning algorithms.

Big data veracity is a HUGE problem! There is no verifying mechanism at all when it comes to social media content being generated in Terra-bytes on the fly. We realized that we have a huge crowd of bored people trying to kill time and lots of data to validate. Piece them together and we can solve this ‘big issue’ in ‘small’ bits.
We crowd source a set of Twitter statuses and ask people to tag them into either ‘happy’, ‘sad’ or ‘neutral’. Each person rates a few statuses. And get lots of people to do the same. And then validate the ratings to see how much they match. And voila! We’re done.

Find the demo here: tinyurl.com/tagmeapp

To verify the ratings, we use a Naive Bayesian classifier running in the back end. This classifies tweets with 83% accuracy. Doesn’t seem like a lot but it’s the best we can got from a machine so far. That’s what our research told us.

Session difficulty level: In-depth talks

Share this session:

Comments

    Leave a Reply