Practical attacks against authorship recognition techniques

Michael Brennan, Rachel Greenstadt

    Research output: Chapter in Book/Report/Conference proceedingConference contribution

    Abstract

    The use of statistical AI techniques in authorship recognition (or stylometry) has contributed to literary and historical breakthroughs. These successes have led to the use of these techniques in criminal investigations and prosecutions. However, few have studied adversarial attacks and their devastating effect on the robustness of existing classification methods. This paper presents a framework for adversarial attacks including obfuscation attacks, where a subject attempts to hide their identity and imitation attacks, where a subject attempts to frame another subject by imitating their writing style. The major contribution of this research is that it demonstrates that both attacks work very well. The obfuscation attack reduces the effectiveness of the techniques to the level of random guessing and the imitation attack succeeds with 68-91% probability depending on the stylometric technique used. These results are made more significant by the fact that the experimental subjects were unfamiliar with stylometric techniques, without specialized knowledge in linguistics, and spent little time on the attacks. This paper also provides another significant contribution to the field in using human subjects to empirically validate the claim of high accuracy for current techniques (without attacks) by reproducing results for three representative stylometric methods.

    Original languageEnglish (US)
    Title of host publicationProceedings of the 21st Innovative Applications of Artificial Intelligence Conference, IAAI-09
    Pages60-65
    Number of pages6
    StatePublished - Dec 1 2009
    Event21st Innovative Applications of Artificial Intelligence Conference, IAAI-09 - Pasadena, CA, United States
    Duration: Jul 14 2009Jul 16 2009

    Publication series

    NameProceedings of the 21st Innovative Applications of Artificial Intelligence Conference, IAAI-09

    Conference

    Conference21st Innovative Applications of Artificial Intelligence Conference, IAAI-09
    CountryUnited States
    CityPasadena, CA
    Period7/14/097/16/09

    Fingerprint

    Linguistics

    ASJC Scopus subject areas

    • Artificial Intelligence
    • Computer Science Applications
    • Computer Vision and Pattern Recognition

    Cite this

    Brennan, M., & Greenstadt, R. (2009). Practical attacks against authorship recognition techniques. In Proceedings of the 21st Innovative Applications of Artificial Intelligence Conference, IAAI-09 (pp. 60-65). (Proceedings of the 21st Innovative Applications of Artificial Intelligence Conference, IAAI-09).

    Practical attacks against authorship recognition techniques. / Brennan, Michael; Greenstadt, Rachel.

    Proceedings of the 21st Innovative Applications of Artificial Intelligence Conference, IAAI-09. 2009. p. 60-65 (Proceedings of the 21st Innovative Applications of Artificial Intelligence Conference, IAAI-09).

    Research output: Chapter in Book/Report/Conference proceedingConference contribution

    Brennan, M & Greenstadt, R 2009, Practical attacks against authorship recognition techniques. in Proceedings of the 21st Innovative Applications of Artificial Intelligence Conference, IAAI-09. Proceedings of the 21st Innovative Applications of Artificial Intelligence Conference, IAAI-09, pp. 60-65, 21st Innovative Applications of Artificial Intelligence Conference, IAAI-09, Pasadena, CA, United States, 7/14/09.
    Brennan M, Greenstadt R. Practical attacks against authorship recognition techniques. In Proceedings of the 21st Innovative Applications of Artificial Intelligence Conference, IAAI-09. 2009. p. 60-65. (Proceedings of the 21st Innovative Applications of Artificial Intelligence Conference, IAAI-09).
    Brennan, Michael ; Greenstadt, Rachel. / Practical attacks against authorship recognition techniques. Proceedings of the 21st Innovative Applications of Artificial Intelligence Conference, IAAI-09. 2009. pp. 60-65 (Proceedings of the 21st Innovative Applications of Artificial Intelligence Conference, IAAI-09).
    @inproceedings{35835e357e4d47ae8d8d4641a737c0e7,
    title = "Practical attacks against authorship recognition techniques",
    abstract = "The use of statistical AI techniques in authorship recognition (or stylometry) has contributed to literary and historical breakthroughs. These successes have led to the use of these techniques in criminal investigations and prosecutions. However, few have studied adversarial attacks and their devastating effect on the robustness of existing classification methods. This paper presents a framework for adversarial attacks including obfuscation attacks, where a subject attempts to hide their identity and imitation attacks, where a subject attempts to frame another subject by imitating their writing style. The major contribution of this research is that it demonstrates that both attacks work very well. The obfuscation attack reduces the effectiveness of the techniques to the level of random guessing and the imitation attack succeeds with 68-91{\%} probability depending on the stylometric technique used. These results are made more significant by the fact that the experimental subjects were unfamiliar with stylometric techniques, without specialized knowledge in linguistics, and spent little time on the attacks. This paper also provides another significant contribution to the field in using human subjects to empirically validate the claim of high accuracy for current techniques (without attacks) by reproducing results for three representative stylometric methods.",
    author = "Michael Brennan and Rachel Greenstadt",
    year = "2009",
    month = "12",
    day = "1",
    language = "English (US)",
    isbn = "9781577354239",
    series = "Proceedings of the 21st Innovative Applications of Artificial Intelligence Conference, IAAI-09",
    pages = "60--65",
    booktitle = "Proceedings of the 21st Innovative Applications of Artificial Intelligence Conference, IAAI-09",

    }

    TY - GEN

    T1 - Practical attacks against authorship recognition techniques

    AU - Brennan, Michael

    AU - Greenstadt, Rachel

    PY - 2009/12/1

    Y1 - 2009/12/1

    N2 - The use of statistical AI techniques in authorship recognition (or stylometry) has contributed to literary and historical breakthroughs. These successes have led to the use of these techniques in criminal investigations and prosecutions. However, few have studied adversarial attacks and their devastating effect on the robustness of existing classification methods. This paper presents a framework for adversarial attacks including obfuscation attacks, where a subject attempts to hide their identity and imitation attacks, where a subject attempts to frame another subject by imitating their writing style. The major contribution of this research is that it demonstrates that both attacks work very well. The obfuscation attack reduces the effectiveness of the techniques to the level of random guessing and the imitation attack succeeds with 68-91% probability depending on the stylometric technique used. These results are made more significant by the fact that the experimental subjects were unfamiliar with stylometric techniques, without specialized knowledge in linguistics, and spent little time on the attacks. This paper also provides another significant contribution to the field in using human subjects to empirically validate the claim of high accuracy for current techniques (without attacks) by reproducing results for three representative stylometric methods.

    AB - The use of statistical AI techniques in authorship recognition (or stylometry) has contributed to literary and historical breakthroughs. These successes have led to the use of these techniques in criminal investigations and prosecutions. However, few have studied adversarial attacks and their devastating effect on the robustness of existing classification methods. This paper presents a framework for adversarial attacks including obfuscation attacks, where a subject attempts to hide their identity and imitation attacks, where a subject attempts to frame another subject by imitating their writing style. The major contribution of this research is that it demonstrates that both attacks work very well. The obfuscation attack reduces the effectiveness of the techniques to the level of random guessing and the imitation attack succeeds with 68-91% probability depending on the stylometric technique used. These results are made more significant by the fact that the experimental subjects were unfamiliar with stylometric techniques, without specialized knowledge in linguistics, and spent little time on the attacks. This paper also provides another significant contribution to the field in using human subjects to empirically validate the claim of high accuracy for current techniques (without attacks) by reproducing results for three representative stylometric methods.

    UR - http://www.scopus.com/inward/record.url?scp=74949095118&partnerID=8YFLogxK

    UR - http://www.scopus.com/inward/citedby.url?scp=74949095118&partnerID=8YFLogxK

    M3 - Conference contribution

    AN - SCOPUS:74949095118

    SN - 9781577354239

    T3 - Proceedings of the 21st Innovative Applications of Artificial Intelligence Conference, IAAI-09

    SP - 60

    EP - 65

    BT - Proceedings of the 21st Innovative Applications of Artificial Intelligence Conference, IAAI-09

    ER -