AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Reducing Sentiment Bias in Pre-trained Sentiment Classification via Adaptive Gumbel Attack

Jiachen Tian, Shizhan Chen, Xiaowang Zhang, Xin Wang, Zhiyong Feng

AAAI 2023 · Label-efficient and robust learning › Recognition and long-tail · approach: Data augmentation and generalization bias · 4 citations (OpenAlex)

Read the paper

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Apply Gater during fine-tuningBERT · SST-2accuracy: 94.9 -> 95.2Improved
Apply Gater during fine-tuningBERT · YELP-5accuracy: 70.6 -> 71.5Improved
Apply Gater during fine-tuningBERT · YELP-2accuracy: 98.1 -> 98.1No clear effect
Apply Gater during fine-tuningBERT · IMDbaccuracy: 95.4 -> 95.8Improved
Apply Gater during fine-tuningBERT · Amazon-5accuracy: 62.2 -> 62.8Improved
Apply Gater during fine-tuningBERT · SemEvalaccuracy: 84.1 -> 86.1Improved
Use excessive numbers of experts instead of adapting the count (ablation)Gater · all datasetsaccuracyHurt
Adapt the number of experts using classification confidence (ablation)Gater · all datasetsaccuracyImproved
Replace the multi-channel parameter update with Adam (ablation)Gater · SST-2accuracyHurt
Replace the multi-channel parameter update with Adam (ablation)Gater · Amazon-5accuracyHurt
Use Gater to debias entity representationsBERT, RoBERTa, ALBERT, and ELECTRA · all seven datasetssentiment scoreImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy