Sarcasm detection in written text, particularly on social media, poses a significant challenge due to its context-dependent nature. Our study introduces a novel hybrid approach using Capsule Neural Networks (CNN) combined with Long Short-Term Memory (LSTM) networks for accurately identifying sarcasm in the Self-Annotated Reddit Corpus (SARC). The study focuses on leveraging advanced feature extraction techniques like Word2Vec and TF-IDF, combined with dimensionality reduction methods such as PCA and LDA, to enhance the model's performance.
The methodology involves preprocessing Reddit comments, including tokenization, stop word removal, and stemming, to prepare the data for deep learning models. Our approach, which combines the strengths of CNN and LSTM, demonstrated high effectiveness in capturing the nuanced context of sarcastic expressions. The Capsule CNN model achieved a remarkable accuracy of 95.60% in distinguishing sarcastic from non-sarcastic comments, outperforming other models like standalone CNN and LSTM.
This research highlights the potential of hybrid neural network models in improving sarcasm detection, contributing to advancements in sentiment analysis and natural language processing. The results are promising for applications in social media analytics, online reputation management, and customer service.
Links:
Full Text: https://www.igminresearch.com/articles/html/igmin137
DOI Link: https://dx.doi.org/10.61927/igmin137
Comments
Post a Comment