Location
https://ccse.kennesaw.edu/computing-showcase/cday-programs/fall23program.php
Document Type
Event
Start Date
30-11-2023 4:00 PM
Description
Sarcasm identification is a vital challenge in natural language processing. In this project, we address this challenge by employing a context-sensitive approach that leverages deep learning, transformer learning, and conventional machine learning models. We conducted our research using two benchmark datasets: Twitter and Internet Argument Corpus (IAC-v2). Our three primary models—Bi-LSTM with GloVe embeddings, BERT, and feature fusion—outperformed baseline methods, achieving an 89.4% highest accuracy on Twitter datasets and an 81.2% highest precision on IAC-v2. These results highlight the effectiveness of our approach in sarcasm detection, with significant implications for sentiment analysis and opinion mining. While our project provides promising results on benchmark datasets, further testing on live tweet datasets is essential to validate its real-world predictive capabilities. This project contributes to the ongoing efforts to enhance communication understanding in the digital era.
Included in
GR-450 Enhancing Sarcasm Detection with Context Sensitivity
https://ccse.kennesaw.edu/computing-showcase/cday-programs/fall23program.php
Sarcasm identification is a vital challenge in natural language processing. In this project, we address this challenge by employing a context-sensitive approach that leverages deep learning, transformer learning, and conventional machine learning models. We conducted our research using two benchmark datasets: Twitter and Internet Argument Corpus (IAC-v2). Our three primary models—Bi-LSTM with GloVe embeddings, BERT, and feature fusion—outperformed baseline methods, achieving an 89.4% highest accuracy on Twitter datasets and an 81.2% highest precision on IAC-v2. These results highlight the effectiveness of our approach in sarcasm detection, with significant implications for sentiment analysis and opinion mining. While our project provides promising results on benchmark datasets, further testing on live tweet datasets is essential to validate its real-world predictive capabilities. This project contributes to the ongoing efforts to enhance communication understanding in the digital era.