Sometimes when you're texting or chatting, words just don't cut it. That's when an emoji steps in to fill the void and save the day. Whether you like it or not, emoji aren't going anywhere — in fact, they've quickly become an alternate way to express an array of emotions. As a result, researchers at MIT's media lab have created DeepMoji, a Twitter-based deep learning algorithm that uses 1.2 billion tweets with emoji to predict the emotion being conveyed by the user via emoji. The researchers believe the more data the model can accumulate, the bigger the possibility of companies or chatbots using the algorithm.
DeepMoji uses deep learning, a subset of machine learning that trains an algorithm to learn and decipher patterns by feeding it huge amounts of data. It's similar to something like QuickDraw from Google, a fun, interactive "game" that has an algorithm try to guess what you draw (which helps grow the database as you use it). In DeepMoji's case, the researchers have given it tweets with emoji so that it can learn which emoji would be matched with a tweet. The team believes that once it can predict what emoji will be used, it'll learn the "emotional content of that sentence." By using real tweets, the algorithm is learning how people actually write and feel, rather than being fed the information.
So far, DeepMoji has figured out when a tweet is being sarcastic or uses slang, as well as understanding how an emoji is being used. For example, a sad emoji can be separated into different categories of emotion like negativity, annoyed, and angry. With other phrases, it knows the difference between something like "this is sh*t" and "this is the sh*t (the former is negative, whereas the latter is positive).
For now, the next steps of the project include the public's help. You can feed the algorithm your last three tweets and emotionally rate them. You can also try out the database for yourself by writing a sentence and seeing if the emoji chosen were correct. The researchers also want to come up with more labels for emotion that range farther out of "positive" and "negative" — so watch the video above to learn more about DeepMoji and how to contribute to the team's research.