Date of Award

12-2024

Document Type

Thesis

Degree Name

Master of Science in Information Systems and Technology

Department

Information and Decision Sciences

First Reader/Committee Chair

Conrad Shayo

Abstract

Sarcasm can be identified in newspaper headlines in digital communication, as it is contextual and has low inter- and intra-observer reliability. This research aims to improve sarcastic comment identification using a natural language processing approach, especially the BI-LSTM. The primary concern is to use fine-tuning methods to enhance the ways that strengthen the sarcasm identification rate, which regards the factors that complicate an automatic identification process. The study explores the impact of techniques such as early stopping, optimal loss function selection, and hyperparameter tuning to enhance the model's performance. The research questions are: Q1) How can fine-tuning techniques for BI-LSTM models be effectively applied to detect sarcasm in varied digital communications? Q2) How do the evaluation metrics, such as the confusion matrix and classification report, improve the accuracy of sarcasm detection? The findings were follows: Q1) for newspaper headline datasets pre-processed and tested through the BI-LSTM model, there was a 7% improvement in the model's accuracy, from 86% to 93%. The conclusion is that application of the BI-LSTM model highlighted the need for fine-tuning procedures to enhance sarcasm prediction. For Q2) The confusion matrix and classification report further supported the analysis, offering a clear understanding of the model’s strengths and weaknesses. Ultimately, these methods led to a significant improvement in accuracy, from 90% to 93%, and demonstrated the model’s capability to accurately detect sarcasm in text. The results extend the state of the art in sarcasm detection by focusing on model optimization, which might be relevant in various NLP applications, such as sentiment analysis, human-like conversation, and social media moderation[CS1] [ST2] . Areas for further study include: (a) the application of forward ductile and robust models, on top of which additional detection might be attempted, and (b) extending the existing modes of communication and embracing additional modes such as images and videos. Moreover, (c), NLP can be used in other languages, not only in English.

Download

  Contact Author

Share

COinS