Task-Oriented Adversarial Attacks for Aspect-Based Sentiment Analysis Models
Adversarial attacks deliberately modify deep learning inputs, mislead models, and cause incorrect results. Previous adversarial attacks on sentiment analysis models have demonstrated success in misleading these models. However, most existing attacks in sentiment analysis have applied a generalized a...
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2025-01-01
|
| Series: | Applied Sciences |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2076-3417/15/2/855 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1832589173658222592 |
|---|---|
| author | Monserrat Vázquez-Hernández Ignacio Algredo-Badillo Luis Villaseñor-Pineda Mariana Lobato-Báez Juan Carlos Lopez-Pimentel Luis Alberto Morales-Rosales |
| author_facet | Monserrat Vázquez-Hernández Ignacio Algredo-Badillo Luis Villaseñor-Pineda Mariana Lobato-Báez Juan Carlos Lopez-Pimentel Luis Alberto Morales-Rosales |
| author_sort | Monserrat Vázquez-Hernández |
| collection | DOAJ |
| description | Adversarial attacks deliberately modify deep learning inputs, mislead models, and cause incorrect results. Previous adversarial attacks on sentiment analysis models have demonstrated success in misleading these models. However, most existing attacks in sentiment analysis have applied a generalized approach to input modifications, without considering the characteristics and objectives of the different analysis levels. Specifically, for aspect-based sentiment analysis, there is a lack of attack methods that modify inputs in accordance with the evaluated aspects. Consequently, unnecessary modifications are made, compromising the input semantics, making the changes more detectable, and avoiding the identification of new vulnerabilities. In previous work, we proposed a model to generate adversarial examples in particular for aspect-based sentiment analysis. In this paper, we assess the effectiveness of our adversarial example model in negatively impacting aspect-based model results while maintaining high levels of semantic inputs. To conduct this evaluation, we propose diverse adversarial attacks across different dataset domains, target architectures, and consider distinct levels of victim model knowledge, thus obtaining a comprehensive evaluation. The obtained results demonstrate that our approach outperforms existing attack methods in terms of accuracy reduction and semantic similarity, achieving a 65.30% reduction in model accuracy with a low perturbation ratio of 7.79%. These findings highlight the importance of considering task-specific characteristics when designing adversarial examples, as even simple modifications to elements that support task classification can successfully mislead models. |
| format | Article |
| id | doaj-art-30e4580ca0a0426a81f448c38fc79765 |
| institution | Kabale University |
| issn | 2076-3417 |
| language | English |
| publishDate | 2025-01-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Applied Sciences |
| spelling | doaj-art-30e4580ca0a0426a81f448c38fc797652025-01-24T13:21:04ZengMDPI AGApplied Sciences2076-34172025-01-0115285510.3390/app15020855Task-Oriented Adversarial Attacks for Aspect-Based Sentiment Analysis ModelsMonserrat Vázquez-Hernández0Ignacio Algredo-Badillo1Luis Villaseñor-Pineda2Mariana Lobato-Báez3Juan Carlos Lopez-Pimentel4Luis Alberto Morales-Rosales5Department of Computer Science, National Institute for Astrophysics, Optics and Electronics, Luis Enrique Erro #1, Sta María Tonanzintla, Puebla 72840, MexicoDepartment of Computer Science, CONAHCYT-National Institute for Astrophysics, Optics and Electronics, Luis Enrique Erro #1, Sta María Tonanzintla, Puebla 72840, MexicoDepartment of Computer Science, National Institute for Astrophysics, Optics and Electronics, Luis Enrique Erro #1, Sta María Tonanzintla, Puebla 72840, MexicoTecnológico Nacional de México (TECNM), Instituto Tecnológico Superior de Libres, Camino Real esq. Camino Cuauhtémoc, Barrio de Tetela, Ciudad de Libres, Puebla 73780, MexicoFacultad de Ingeniería, Universidad Panamericana, Álvaro del Portillo 49, Zapopan 45010, MexicoFacultad de Ingeniería Civil, CONACYT-Universidad Michoacana de San Nicolás de Hidalgo, C. de Santiago Tapia 403, Centro, Morelia 58000, MexicoAdversarial attacks deliberately modify deep learning inputs, mislead models, and cause incorrect results. Previous adversarial attacks on sentiment analysis models have demonstrated success in misleading these models. However, most existing attacks in sentiment analysis have applied a generalized approach to input modifications, without considering the characteristics and objectives of the different analysis levels. Specifically, for aspect-based sentiment analysis, there is a lack of attack methods that modify inputs in accordance with the evaluated aspects. Consequently, unnecessary modifications are made, compromising the input semantics, making the changes more detectable, and avoiding the identification of new vulnerabilities. In previous work, we proposed a model to generate adversarial examples in particular for aspect-based sentiment analysis. In this paper, we assess the effectiveness of our adversarial example model in negatively impacting aspect-based model results while maintaining high levels of semantic inputs. To conduct this evaluation, we propose diverse adversarial attacks across different dataset domains, target architectures, and consider distinct levels of victim model knowledge, thus obtaining a comprehensive evaluation. The obtained results demonstrate that our approach outperforms existing attack methods in terms of accuracy reduction and semantic similarity, achieving a 65.30% reduction in model accuracy with a low perturbation ratio of 7.79%. These findings highlight the importance of considering task-specific characteristics when designing adversarial examples, as even simple modifications to elements that support task classification can successfully mislead models.https://www.mdpi.com/2076-3417/15/2/855adversarial attacksdeep learningvulnerabilitiesaspect-basedsentiment analysis |
| spellingShingle | Monserrat Vázquez-Hernández Ignacio Algredo-Badillo Luis Villaseñor-Pineda Mariana Lobato-Báez Juan Carlos Lopez-Pimentel Luis Alberto Morales-Rosales Task-Oriented Adversarial Attacks for Aspect-Based Sentiment Analysis Models Applied Sciences adversarial attacks deep learning vulnerabilities aspect-based sentiment analysis |
| title | Task-Oriented Adversarial Attacks for Aspect-Based Sentiment Analysis Models |
| title_full | Task-Oriented Adversarial Attacks for Aspect-Based Sentiment Analysis Models |
| title_fullStr | Task-Oriented Adversarial Attacks for Aspect-Based Sentiment Analysis Models |
| title_full_unstemmed | Task-Oriented Adversarial Attacks for Aspect-Based Sentiment Analysis Models |
| title_short | Task-Oriented Adversarial Attacks for Aspect-Based Sentiment Analysis Models |
| title_sort | task oriented adversarial attacks for aspect based sentiment analysis models |
| topic | adversarial attacks deep learning vulnerabilities aspect-based sentiment analysis |
| url | https://www.mdpi.com/2076-3417/15/2/855 |
| work_keys_str_mv | AT monserratvazquezhernandez taskorientedadversarialattacksforaspectbasedsentimentanalysismodels AT ignacioalgredobadillo taskorientedadversarialattacksforaspectbasedsentimentanalysismodels AT luisvillasenorpineda taskorientedadversarialattacksforaspectbasedsentimentanalysismodels AT marianalobatobaez taskorientedadversarialattacksforaspectbasedsentimentanalysismodels AT juancarloslopezpimentel taskorientedadversarialattacksforaspectbasedsentimentanalysismodels AT luisalbertomoralesrosales taskorientedadversarialattacksforaspectbasedsentimentanalysismodels |