Detection emotions in text
It’s common to hear about detection of emotions in pictures of people faces, but other area that can be usefull to now the emotions of people with only analysing what they write in places like comments in Facebook or Twiter. This open a new possibility to automatize the analisys of the emotions and impact of the users about a product, a service or even the opinion of the goverment.
Understanding the user is so important to many sectors, even in economics it can be a main factor to predict the success in the market or the risk that a company could face in a future.
In this case, using the method of the Mechanical Turk for its low cost and scalability, it tries to classify short informal english messages in fice classes: happy, angry, sad, excited and fear.
The plan is to identify “how much” of a emotion that messages could be interpreted. They use a private dataset of more than 780,000 messages that 60% used in trainig, 20% in validation and 20% for testing; in a experiment of a three layer ANN with 125, 25, 5 nodes.
The funny part of the results, its to see that the more accurated predictions are in negative emotions, except in fear that is the worst one.
