Adversarial Text Perturbation

Adversarial text perturbation explores how small changes to text can drastically alter the output of natural language processing (NLP) models, aiming to understand and mitigate this vulnerability. Current research focuses on developing both attacks (methods to create these perturbations) and defenses, often employing transformer models like BERT and RoBERTa, and exploring techniques like latent representation randomization and data augmentation strategies. This research is crucial for improving the robustness and reliability of NLP systems across various applications, from sentiment analysis in news to content moderation on social media, where susceptibility to adversarial attacks can have significant consequences.

Papers

February 18, 2024

A Curious Case of Searching for the Correlation between Training Data and Adversarial Robustness of Transformer Textual Models
Cuong Dang, Dung D. Le, Thai Le
Training Data Adversarial Training Adversarial Robustness Search Query Total Correlation Adversarial Evaluation Adversarial Text Perturbation

February 3, 2024

Analyzing Sentiment Polarity Reduction in News Presentation through Contextual Perturbation and Large Language Models
Alapan Kuila, Somnath Jena, Sudeshna Sarkar, Partha Pratim Chakrabarti
User Sentiment Sentiment Polarity News Article Sentiment Information Sentiment Bias Context Perturbation Adversarial Text Perturbation

October 2, 2023

Fooling the Textual Fooler via Randomizing Latent Representations
Duy C. Hoang, Quang H. Nguyen, Saurav Manchanda, MinLong Peng, Kok-Seng Wong, Khoa D. Doan
Adversarial Attack Adversarial Example Adversarial Perturbation Latent Representation Word Level Adversarial Adversarial Text Perturbation

May 16, 2023

Adversarial Word Dilution as Text Data Augmentation in Low-Resource Regime
Junfan Chen, Richong Zhang, Zheyan Luo, Chunming Hu, Yongyi Mao
Data Augmentation Low Resource Effective Data Augmentation Text Data Augmentation Adversarial Text Perturbation

June 23, 2022

BERT Rankers are Brittle: a Study using Adversarial Document Perturbations
Yumeng Wang, Lijun Lyu, Avishek Anand
Adversarial Attack Study Feature BERT Based Adversarial Input Token Embeddings Brittle Fracture Adversarial Text Perturbation

February 19, 2022

Data-Driven Mitigation of Adversarial Text Perturbation
Rasika Bhalerao, Mohammad Al-Rubaie, Anand Bhaskar, Igor Markov
Word Embeddings Obfuscation Technique Adaptive Mitigation Adversarial Text Perturbation