Enhancing Contextual Understanding in Sentiment Analysis Using Advanced Transformer Architectures.
ID：144
Submission ID：589 View Protection：ATTENDEE
Updated Time：2026-07-25 16:44:07 Hits：16
Online
Start Time：2026-07-31 14:40 （Asia/Kolkata）
Duration：15min
Session：[S7] Disruptive Technologies for Manufacturing » [S7-2] Disruptive Technologies for Manufacturing
Video
No Permission
Presentation File
Tips: The file permissions under this presentation are only for participants. You have not logged in yet and cannot view it temporarily.
Abstract
Transformer-based models have improved sentiment analysis greatly, but it is difficult to detect contextual complexities, including sarcasm, irony, and implicit sentiment. This work aims to solve this problem by assessing state-of-the-art transformer designs and suggesting an ensemble architecture to provide a better understanding of the context. They are initially evaluated on baseline models like BERT and RoBERTa and then advanced models like T5 and ELECTRA are used in order to enhance context-aware representation. Various benchmark datasets were experimented with, such as IMDB, Twitter, and a sarcasm dataset, to take into account a variety of linguistic properties. Accuracy and F1-score were used as measures of performance, and the context-sensitive sentiment detection was specifically considered. The findings reveal that although more sophisticated models are better than the basic models, the proposed ensemble framework is always superior in all datasets. In particular, it shows better ability to recognize subtle sentiment expression, particularly in contexts with sarcasm. In general, the work indicates the importance of combining various transformer models to improve the contextual knowledge and gives a viable way forward in designing an effective sentiment analysis system.
Keywords
Sentiment Analysis, Transformer Models, Contextual Understanding, Ensemble Learning, Sarcasm Detection
Speaker
Comment submit