Additive Feature Attribution Explainable Methods to Craft Adversarial Attacks for Text Classification and Text Regression Article Swipe

MAKE IMPACT

Yidong Chai , Ruicheng Liang , Sagar Samtani , Hongyi Zhu , Meng Wang , Yezheng Liu , Yuanchun Jiang · YOU? · · 2022 · Open Access · · DOI: https://doi.org/10.36227/techrxiv.17185568.v2

Deep learning (DL) models have significantly improved the performance of text classification and text regression tasks. However, DL models are often strikingly vulnerable to adversarial attacks. Many researchers have aimed to develop adversarial attacks against DL models in realistic black-box settings (i.e., assumes no model knowledge is accessible to attackers) that typically operate with a two-phase framework: (1) sensitivity estimation through gradient-based or deletion-based methods to evaluate the sensitivity of each token to the prediction of the target model, and (2) perturbation execution to craft adversarial examples based on the estimated token sensitivity. However, gradient-based and deletion-based methods used to estimate sensitivity often face issues of capturing the directionality of tokens and overlapping token sensitivities, respectively. In this study, we propose a novel eXplanation-based method for Adversarial Text Attacks (XATA) that leverages additive feature attribution explainable methods, namely LIME or SHAP, to measure the sensitivity of input tokens when crafting black-box adversarial attacks on DL models performing text classification or text regression. We evaluated XATA’s attack performance on DL models executing text classification on three datasets (IMDB Movie Review, Yelp Reviews-Polarity, and Amazon Reviews-Polarity) and DL models conducting text regression on three datasets (My Personality, Drug Review, and CommonLit Readability). The proposed XATA outperformed the existing gradient-based and deletion-based adversarial attack baselines in both tasks. These findings indicate that the ever-growing research focused on improving the explainability of DL models with additive feature attribution explainable methods can provide attackers with weapons to launch targeted adversarial attacks.

Type

preprint

Language

en

Landing Page

https://doi.org/10.36227/techrxiv.17185568.v2

OA Status

gold

References

53

Related Works

10

OpenAlex ID

https://openalex.org/W4281557628

Raw OpenAlex JSON

OpenAlex ID

https://openalex.org/W4281557628

Canonical identifier for this work in OpenAlex

DOI

https://doi.org/10.36227/techrxiv.17185568.v2

Digital Object Identifier

Title

Additive Feature Attribution Explainable Methods to Craft Adversarial Attacks for Text Classification and Text Regression

Work title

Type

preprint

OpenAlex work type

Language

en

Primary language

Publication year

2022

Year of publication

Publication date

2022-05-25

Full publication date if available

Authors

Yidong Chai, Ruicheng Liang, Sagar Samtani, Hongyi Zhu, Meng Wang, Yezheng Liu, Yuanchun Jiang

List of authors in order

Landing page

https://doi.org/10.36227/techrxiv.17185568.v2

Publisher landing page

Open access

Yes

Whether a free full text is available

OA status

gold

Open access status per OpenAlex

OA URL

https://doi.org/10.36227/techrxiv.17185568.v2

Direct OA link when available

Concepts

Computer science, Security token, Adversarial system, Artificial intelligence, Regression, Machine learning, Feature (linguistics), Deep learning, Sensitivity (control systems), Natural language processing, Data mining, Statistics, Mathematics, Engineering, Electronic engineering, Computer security, Philosophy, Linguistics

Top concepts (fields/topics) attached by OpenAlex

Cited by

0

Total citation count in OpenAlex

References (count)

53

Number of works referenced by this work

Related works (count)

10

Other works algorithmically related by OpenAlex

Full payload