Interpretable Multi-Head Self-Attention Architecture for Sarcasm Detection

Technology #2021-113

Key Points

  • An interpretable deep learning model for detecting sarcasm in social media text
  • Invention module aids in identifying crucial sarcastic cue-words from the input while recurrent units learn long-range dependencies between the cue-words to classify the input text better
  • Applicable to marketing research, opinion mining, information classification


The University of Central Florida invention is an interpretable deep learning model to detect sarcasm within social media text. Since social media enables businesses to advertise their products, build brand value, and reach out to their customers, businesses need to process customer feedback in posts and tweets. Sentiment analysis identifies the emotion, either positive, negative, or neutral, associated with these social media texts. The presence of sarcasm in texts is the main hindrance to the performance of sentiment analysis. This invention assists with the identification of sarcastic cues to train text classification systems.


  • Provides the complexity needed to decode sarcasm in text
  • Enables the interpretability of what the model learned to perform its task

Market Application

  • Customer service
  • Marketing research, opinion mining, information classification