Latent-Optimized Adversarial Neural Transfer for Sarcasm Detection

Abstract

The existence of multiple datasets for sarcasm detection prompts us to apply transfer learning to exploit their commonality. The adversarial neural transfer (ANT) framework utilizes multiple loss terms that encourage the source-domain and the target-domain feature distributions to be similar while optimizing for domain-specific performance. However, these objectives may be in conflict, which can lead to optimization difficulties and sometimes diminished transfer. We propose a generalized latent optimization strategy that allows different losses to accommodate each other and improves training dynamics. The proposed method outperforms transfer learning and meta-learning baselines. In particular, we achieve 10.02% absolute performance gain over the previous state of the art on the iSarcasm dataset.

UTNLP at SemEval-2022 Task 6: A Comparative Analysis of Sarcasm Detection using generative-based and mutation-based data augmentation

“…We got an F1-sarcastic of 0.323 using BERT. We achieved our best result with RoBERTa with an F1-sarcastic of 0.414, which was better than LOANT (Guo et al, 2021) model on the same dataset.…”

“…In (Guo et al, 2021), the Latent Optimized Adversarial Neural Transfer (LOANT) model was suggested as a novel latent-optimized adversarial neural transfer model for cross-domain sarcasm detection. LOANT surpasses classical adversarial neural transfer, multitask learning, and meta-learning baselines using stochastic gradient descent (SGD) with a one-step look-ahead and sets a new state-ofthe-art F-score of 0.4101 on the iSarcasm dataset.…”