The algorithm was developed by a team from MIT. In an article for the MIT Technology Review, the researchers explained how emoji provides a convenient starting point for the AI. Since emoji already acts as a kind of “labelling system” for emotional content, the presence of certain emoji in a tweet is a good indicator of its overall sentiment.
The neural network uses emoji to assess tweets for general emotion. It then springboards from the initial analysis to start looking for elements of sarcasm. In a trial, the emoji-trained algorithm proved to be “far better” at detecting sarcasm than a comparable one without the training.
“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, ” said MIT Media associate professor Iyad Rahwan. “The neural network learned the connection between a certain kind of language and an emoji.”
READ NEXT: Microsoft sees cyberattacks as part of the fabric of society
The researchers collected a dataset of over 55 billion tweets for the experiment. They then proceeded to train the algorithm with a subset of 1.2 billion tweets containing various combinations of 64 popular emoji. The test was conducted with “several” benchmarks previously shown to be effective at surfacing sentiment indicators in text.
Although the emoji-trained algorithm proved to be effective, experts are sceptical it could have practical applications. Gary King, director of the Institute for Quantitative Social Science at Harvard University, told the MIT Technology Review that the detected sarcasm “really doesn’t matter” if it’s too nuanced for a human reader to appreciate. Rahwan also acknowledged that the benchmarks used “don’t capture” the subtleties of sentiment that humans can recognise.
READ NEXT: Augmented and virtual reality spending to double each year
Even so, the AI-powered future that tech visionaries are pitching is likely to require machines that can sense human emotions. Training these neural networks to understand sentiment will be one of the biggest hurdles in bringing them online. The team’s research demonstrates that emoji-based analysis of existing material could be an effective way to improve the accuracy of text-based algorithms.
“If machines are going to cooperate with us, then they’re going to have to understand us, and emotion is really hard,” Rahwan said.
Rahwan and his fellow researchers have created a website that lets you assist the algorithm, known as DeepMoji, in furthering its training. You can upload your own messages to have the algorithm analyse the sentiment expressed in them. The service will return the emotions as emoji that best represent the sentence.
