DeepMoji is an artificial intelligence created by MIT and trained to associate emotional emojis with a message. This new AI would be able to recognize happy, sad, or sarcastic messages as you type them – and then associate the words with an appropriate emoji.
Understanding the intricacies of a message is not always obvious to an AI. Especially when it comes to adding emotions. This challenge, however, has been embarked upon by researchers from the prestigious Massachusetts Institute of Technology.
DeepMoji is trained by using more than 1.2 billion tweets that were screened from 55 billion tweets. Thanks to this, the AI can predict which emoji will fit best with a sentence. This new AI is trained to go far beyond the current emotional emoji related works.
This is a model that groups emotional emojis into larger categories initially. For instance, emojis could be grouped into emotions such as positivity, negativity, and love. This AI does not stop here. It goes further to subcategorize even emojis in the same emotional category, thus narrowing down the lane and making things more specific. For instance, though ‘annoyed’ and ‘angry’ can be in the same negativity category, the AI learns to put these into two different subcategories.
Even though it would narrow down to make things more specific, the model will, however, display a degree of certainty about this association, as it depends on an assortment of words and punctuation. For instance, the “…” being often associated with sarcasm.
The direct utility of such an intelligence is rather limited, especially if viewed from the point of an ordinary person. On the other hand, it could very well be associated with social listening tools to help brands in order to better understand the meaning of a customer’s message.
The model is surely a progress when it comes to the development of emotion analysis software. It would, on the basics, help take emotional analysis from the current positive and negative analysis to include all possibilities in the future.
You can try the tool directly here