A new AI tool funded, in part, by the U.S. military is proven helpful at an almost impossible task. Sarcasm isn’t always easy to recognize in conversation, it’s pretty difficult for a computer program to do it. This new tool can help intelligence agencies to better apply AI to their investigations by forgoing social media posts that aren’t serious.
The method banks on what the researchers relate to as a self-attention architecture – a method used for training complex artificial intelligence programs to give more importance to some words over others.
The researchers’ AI concentrates on disparities between text and imagery and then integrates those results to make predictions. It also relates hashtags to tweet text to help evaluate the sentiment a user is trying to articulate.
Researchers mapped out how some keywords relate to other words. “For example, words such as ‘just’, ‘again’, ‘totally’, ‘!’, have darker edges linking them with every other word in a sentence. These are the words in the sentence that suggest sarcasm and gather higher attention than others.
“Sarcasm isn’t always easy to identify in conversation, so you can understand it’s pretty challenging for a computer program to do it. We developed an interpretable deep learning model using multi-head self-attention and gated recurrent units. The multi-head self-attention module assists in specifying crucial sarcastic cue-words from the input, and the recurrent units learn long-range reliances between these cue-words to better classify the input text.” says Assistant Professor of engineering, Ivan Garibay
It’s not the first time researchers have tried to use machine learning or artificial intelligence to catch the sarcasm from text, like social media posts. Many of which were based on training algorithms that hunt for many specific clues picked by the researchers, like words indicating particular emotions or even emojis.
Source: Tech Juice