EUVIMEDEUVIMED entdecken
Crossref

A Design of Multimodal Sarcasm Detection Integrating Multi-View Interactive Model and Dual-Branch mBERT with LSTM Model

Thanveer Jahan, E Balakrishna, Latha Panjala, Swetha Arra, Nagunuri Rajendar, K Rajasekhar, Salim Amirali Jiwani

Journal of Computer Allied Intelligence · 30.04.2026 · 4 (2) · 17-40

DOI: 10.69996/jcai.2026008 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Computer Allied Intelligence, online erschienen am 30.04.2026. Verfasst von Thanveer Jahan und 6 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Multimodal Sarcasm Detection integration enhances cross-modal alignment by assigning importance weights and promoting interactions among textual, visual, and knowledge-based features. Combining GEN from HuffPost, The Onion, and Sarcasm Corpus V2 helped us accurately represent both formal and sarcastic ways of speaking. To tackle this problem a new novel has developed the MSDMDL model, using DenseNet-121 to extract image features and BERT for text and knowledge representation, thus enabling a deeper semantic encoding across different types of data. The Multi-View Interaction Module aligns text, image, and knowledge features through attention mechanisms for better cross-modal interactions. The Fusion Module enhances features and applies multi-scale convolution to not only detect fine details but also overall patterns, therefore improving multimodal understanding. The Feature Integration Module refines fused features through the use of fully connected layers, normalization, and dropout, ultimately resulting in a better ability to generalize and robustness. Besides, the LSTM Module is responsible for learning the order and dependencies within text data. In the meantime, an attention mechanism is used to assign weights to different words based on their importance for identifying sarcasm. The Dual-Branch mBERT Module derives additional semantic information through supervised attention and cross-lingual learning, thus increasing its capability to work across different languages and domains. The performance analysis is accuracy, loss, precision, f1-score, recall, and ROC curve.

Zitieren

APA

Jahan, T., Balakrishna, E., Panjala, L., Arra, S., Rajendar, N., Rajasekhar, K., & Jiwani, S. A. (2026). A Design of Multimodal Sarcasm Detection Integrating Multi-View Interactive Model and Dual-Branch mBERT with LSTM Model. Journal of Computer Allied Intelligence, 4(2), 17-40. https://doi.org/10.69996/jcai.2026008

Vancouver

Jahan T, Balakrishna E, Panjala L, Arra S, Rajendar N, Rajasekhar K, et al. A Design of Multimodal Sarcasm Detection Integrating Multi-View Interactive Model and Dual-Branch mBERT with LSTM Model. Journal of Computer Allied Intelligence. 2026;4(2):17-40. doi:10.69996/jcai.2026008

BibTeX
@article{Jahan2026,
  author = {Jahan, Thanveer and Balakrishna, E and Panjala, Latha and Arra, Swetha and Rajendar, Nagunuri and Rajasekhar, K and Jiwani, Salim Amirali},
  title = {A Design of Multimodal Sarcasm Detection Integrating Multi-View Interactive Model and Dual-Branch mBERT with LSTM Model},
  journal = {Journal of Computer Allied Intelligence},
  year = {2026},
  volume = {4},
  number = {2},
  pages = {17--40},
  doi = {10.69996/jcai.2026008}
}
Kostenlos starten →