Identifying Adversarial Attacks on Text Classifiers

TL;DR

This paper creates a large dataset of adversarial text attacks, develops classifiers to detect and identify these attacks, and explores features that improve attack forensics for text classifiers.

Contribution

It introduces an extensive dataset of attack instances, benchmarks classifiers for attack detection and identification, and evaluates feature types for attack forensics.

Findings

Text properties effectively distinguish attacked texts.

Language model features improve attack identification.

Target model features reveal attack influence.

Abstract

The landscape of adversarial attacks against text classifiers continues to grow, with new attacks developed every year and many of them available in standard toolkits, such as TextAttack and OpenAttack. In response, there is a growing body of work on robust learning, which reduces vulnerability to these attacks, though sometimes at a high cost in compute time or accuracy. In this paper, we take an alternate approach -- we attempt to understand the attacker by analyzing adversarial text to determine which methods were used to create it. Our first contribution is an extensive dataset for attack detection and labeling: 1.5~million attack instances, generated by twelve adversarial attacks targeting three classifiers trained on six source datasets for sentiment analysis and abuse detection in English. As our second contribution, we use this dataset to develop and benchmark a number of classifiers for attack identification -- determining if a given text has been adversarially manipulated and by which attack. As a third contribution, we demonstrate the effectiveness of three classes of features for these tasks: text properties, capturing content and presentation of text; language model properties, determining which tokens are more or less probable throughout the input; and target model properties, representing how the text classifier is influenced by the attack, including internal node activations. Overall, this represents a first step towards forensics for adversarial attacks against text classifiers.