Emojis can now help in detecting sarcasm on Twitter!
Millions use emojis every day; they help inject emotions into our often brief and dull online conversations. However, in online communications, it is hard to identify whether a remark is sincere or something made in jest - sarcasm often goes over someone's head! So, MIT researchers have now developed an algorithm named 'DeepMoji' to analyze tweets and detect sarcasm better than humans. Read more!
MIT scientists used popular emojis to train AI system
We often come across sarcastic comments on Twitter, though most of them escape our comprehension. So, MIT Media Associate Professor Iyad Rahwan and his student, Bjarke Felbo, have developed DeepMoji. It can spot sarcasm and emotional subtext on Twitter; its performance is better compared to humans'. DeepMoji could also help social media networks detect and remove online hate speech and abuse.
MIT Associate Professor Iyad Rahwan's statement
Rahwan said: "Because we can't use intonation in our voice or body language to contextualize what we are saying, emoji are the way we do it online. The neural network learned the connection between a certain kind of language and an emoji."
How was the system developed and trained?
Researchers initially wanted to develop a racism-detecting system but realized that understanding of sarcasm is necessary. DeepMoji uses deep learning; it was trained using 1.2 billion tweets containing 64 popular emojis. It was trained to predict which emoji would be used for a particular message depending on whether it was a sad, happy, or humorous one. It was later taught to identify sarcasm.
Algorithm to be released for anyone to use
DeepMoji's ability to sense emotion and sentiment in texts were tested against various benchmarks; it performed better than other existing systems. It was tested against crowd-sourcing platform Mechanical Turk's volunteers. Its sarcasm detection was 82% accurate compared to an average 76% score of the volunteers.
Machines are going to have to understand us: Rahwan
DeepMoji can track customers' attitudes towards products/brands and identify signals indicating market trends. Also, its ability to analyze tweets can help researchers know how information and influence flow through Twitter. Its emotion-sensing ability could be fundamental to human-to-machine communication. It can aid computers in interpreting human emotions; especially in having robots at workplaces that understand employees' frustration or identify their sarcastic comments.
The DeepMoji Website
MIT researchers created DeepMoji website (deepmoji.mit.edu) for demonstrating the system, which can automatically include suitable emojis into the entered text. The site allows users to annotate their tweets with emotions, contributing to the research. Rahwan says this contribution is an essential element of the research.