How did the University of Central Florida Develop a Sarcasm Detector?

No second thoughts about the fact that how critical has social media become a part of our lives. We rely on social media so much that today imagining a life without it does not sink in. No wonder social media is considered to be one of the best platforms to market and sell different products and services in addition to being a dominant form of communication. With this platform, you stand a chance to reach out to the maximum lot. While this medium is used for driving sales, another area that it caters to is how are our customers reacting to what you are delivering. This is also a platform with the help of which one can address customer queries, properly understand and respond to their feedback as well. All this plays a pivotal role in achieving success.

With the advancements in technology, we are at a stage wherein it is very much possible to understand what the person just next to you feels about the product or service that you are marketing. This is nothing but sentiment analysis. Sentiment analysis, simply put, is the automated process of identifying the emotion. The emotion could be anything – it could be positive, negative and even neutral. All this is associated with the text. In other words, sentiment analysis revolves around correctly identifying emotional communication. One such innovative step taken in this aspect is the ability to detect sarcasm in a social media text.

Computer science researchers at the University of Central Florida (UCF) grabbed eyeballs by being able to develop a technique that accurately detects sarcasm in a social media text. This achievement made headlines as their findings were recently published in the journal, Entropy.

Now, here is what the team did. The team-taught a computer model to find patterns that often indicate sarcasm. The program is taught to correctly pick out cue words in sequences that were more likely to indicate sarcasm. Wondering how did the researchers achieve this? Well, they fed the program with large data sets and then checked its accuracy.

According to Ivan Garibay, Assistant Professor of Engineering, the main hindrance in the performance of sentiment analysis is the presence of sarcasm in text. He added that it is not that easy a task to identify sarcasm in a conversation and therefore quite challenging for a computer program to do it as well. He said that the team developed an interpretable deep learning model using multi-head self-attention and gated recurrent units. Here, the multi-head self-attention module caters to identifying crucial sarcastic cue-words from the input. The objective of the gated recurrent units is to learn long-range dependencies between these cue-words to better classify the input text. One of the team members, Ramya Akula, a computer science doctoral student has been working on this under a DARPA grant that supports the organization's Computational Simulation of Online Social Behavior program.

Brian Kettler, a program manager in DARPA's Information Innovation Office (I2O) believes that sarcasm is a major obstacle when it comes to increasing the accuracy of sentiment analysis. This becomes all the more critical when the platform under consideration is social media. He explained the reason for this as sarcasm relying heavily on vocal tones, facial expressions and gestures that cannot be represented in the text. He further noted that recognizing sarcasm in textual online communication is not an easy task as none of these cues are readily available.

The same issue is considered to be one of the prominent challenges that Garibay's Complex Adaptive Systems Lab (CASL) is studying. CASL is an interdisciplinary research group and the researchers are studying a range of problems using data science, machine learning, deep learning, network science, complexity science, cognitive science, social sciences, etc.