Stay on-topic: Generating context-specific fake restaurant reviews

Mika Juuti, Bo Sun, Tatsuya Mori, N. Asokan

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

    2 Citations (Scopus)

    Abstract

    Automatically generated fake restaurant reviews are a threat to online review systems. Recent research has shown that users have difficulties in detecting machine-generated fake reviews hiding among real restaurant reviews. The method used in this work (char-LSTM) has one drawback: it has difficulties staying in context, i.e. when it generates a review for specific target entity, the resulting review may contain phrases that are unrelated to the target, thus increasing its detectability. In this work, we present and evaluate a more sophisticated technique based on neural machine translation (NMT) with which we can generate reviews that stay on-topic. We test multiple variants of our technique using native English speakers on Amazon Mechanical Turk. We demonstrate that reviews generated by the best variant have almost optimal undetectability (class-averaged F-score 47%). We conduct a user study with experienced users and show that our method evades detection more frequently compared to the state-of-the-art (average evasion 3.2/4 vs 1.5/4) with statistical significance, at level α1% (Sect. 4.3). We develop very effective detection tools and reach average F-score of 97% in classifying these. Although fake reviews are very effective in fooling people, effective automatic detection is still feasible.

    Original languageEnglish
    Title of host publicationComputer Security - 23rd European Symposium on Research in Computer Security, ESORICS 2018, Proceedings
    EditorsJavier Lopez, Jianying Zhou, Miguel Soriano
    PublisherSpringer-Verlag
    Pages132-151
    Number of pages20
    ISBN (Print)9783319990729
    DOIs
    Publication statusPublished - 2018 Jan 1
    Event23rd European Symposium on Research in Computer Security, ESORICS 2018 - Barcelona, Spain
    Duration: 2018 Sep 32018 Sep 7

    Publication series

    NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
    Volume11098 LNCS
    ISSN (Print)0302-9743
    ISSN (Electronic)1611-3349

    Other

    Other23rd European Symposium on Research in Computer Security, ESORICS 2018
    CountrySpain
    CityBarcelona
    Period18/9/318/9/7

    Fingerprint

    Context
    Review
    Multiple Tests
    Machine Translation
    Target
    Detectability
    User Studies
    Statistical Significance
    Evaluate
    Demonstrate
    Class

    ASJC Scopus subject areas

    • Theoretical Computer Science
    • Computer Science(all)

    Cite this

    Juuti, M., Sun, B., Mori, T., & Asokan, N. (2018). Stay on-topic: Generating context-specific fake restaurant reviews. In J. Lopez, J. Zhou, & M. Soriano (Eds.), Computer Security - 23rd European Symposium on Research in Computer Security, ESORICS 2018, Proceedings (pp. 132-151). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11098 LNCS). Springer-Verlag. https://doi.org/10.1007/978-3-319-99073-6_7

    Stay on-topic : Generating context-specific fake restaurant reviews. / Juuti, Mika; Sun, Bo; Mori, Tatsuya; Asokan, N.

    Computer Security - 23rd European Symposium on Research in Computer Security, ESORICS 2018, Proceedings. ed. / Javier Lopez; Jianying Zhou; Miguel Soriano. Springer-Verlag, 2018. p. 132-151 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11098 LNCS).

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

    Juuti, M, Sun, B, Mori, T & Asokan, N 2018, Stay on-topic: Generating context-specific fake restaurant reviews. in J Lopez, J Zhou & M Soriano (eds), Computer Security - 23rd European Symposium on Research in Computer Security, ESORICS 2018, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 11098 LNCS, Springer-Verlag, pp. 132-151, 23rd European Symposium on Research in Computer Security, ESORICS 2018, Barcelona, Spain, 18/9/3. https://doi.org/10.1007/978-3-319-99073-6_7
    Juuti M, Sun B, Mori T, Asokan N. Stay on-topic: Generating context-specific fake restaurant reviews. In Lopez J, Zhou J, Soriano M, editors, Computer Security - 23rd European Symposium on Research in Computer Security, ESORICS 2018, Proceedings. Springer-Verlag. 2018. p. 132-151. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). https://doi.org/10.1007/978-3-319-99073-6_7
    Juuti, Mika ; Sun, Bo ; Mori, Tatsuya ; Asokan, N. / Stay on-topic : Generating context-specific fake restaurant reviews. Computer Security - 23rd European Symposium on Research in Computer Security, ESORICS 2018, Proceedings. editor / Javier Lopez ; Jianying Zhou ; Miguel Soriano. Springer-Verlag, 2018. pp. 132-151 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).
    @inproceedings{7b61488ddbf64b1a991ec751fb1f9fa3,
    title = "Stay on-topic: Generating context-specific fake restaurant reviews",
    abstract = "Automatically generated fake restaurant reviews are a threat to online review systems. Recent research has shown that users have difficulties in detecting machine-generated fake reviews hiding among real restaurant reviews. The method used in this work (char-LSTM) has one drawback: it has difficulties staying in context, i.e. when it generates a review for specific target entity, the resulting review may contain phrases that are unrelated to the target, thus increasing its detectability. In this work, we present and evaluate a more sophisticated technique based on neural machine translation (NMT) with which we can generate reviews that stay on-topic. We test multiple variants of our technique using native English speakers on Amazon Mechanical Turk. We demonstrate that reviews generated by the best variant have almost optimal undetectability (class-averaged F-score 47{\%}). We conduct a user study with experienced users and show that our method evades detection more frequently compared to the state-of-the-art (average evasion 3.2/4 vs 1.5/4) with statistical significance, at level α1{\%} (Sect. 4.3). We develop very effective detection tools and reach average F-score of 97{\%} in classifying these. Although fake reviews are very effective in fooling people, effective automatic detection is still feasible.",
    author = "Mika Juuti and Bo Sun and Tatsuya Mori and N. Asokan",
    year = "2018",
    month = "1",
    day = "1",
    doi = "10.1007/978-3-319-99073-6_7",
    language = "English",
    isbn = "9783319990729",
    series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
    publisher = "Springer-Verlag",
    pages = "132--151",
    editor = "Javier Lopez and Jianying Zhou and Miguel Soriano",
    booktitle = "Computer Security - 23rd European Symposium on Research in Computer Security, ESORICS 2018, Proceedings",

    }

    TY - GEN

    T1 - Stay on-topic

    T2 - Generating context-specific fake restaurant reviews

    AU - Juuti, Mika

    AU - Sun, Bo

    AU - Mori, Tatsuya

    AU - Asokan, N.

    PY - 2018/1/1

    Y1 - 2018/1/1

    N2 - Automatically generated fake restaurant reviews are a threat to online review systems. Recent research has shown that users have difficulties in detecting machine-generated fake reviews hiding among real restaurant reviews. The method used in this work (char-LSTM) has one drawback: it has difficulties staying in context, i.e. when it generates a review for specific target entity, the resulting review may contain phrases that are unrelated to the target, thus increasing its detectability. In this work, we present and evaluate a more sophisticated technique based on neural machine translation (NMT) with which we can generate reviews that stay on-topic. We test multiple variants of our technique using native English speakers on Amazon Mechanical Turk. We demonstrate that reviews generated by the best variant have almost optimal undetectability (class-averaged F-score 47%). We conduct a user study with experienced users and show that our method evades detection more frequently compared to the state-of-the-art (average evasion 3.2/4 vs 1.5/4) with statistical significance, at level α1% (Sect. 4.3). We develop very effective detection tools and reach average F-score of 97% in classifying these. Although fake reviews are very effective in fooling people, effective automatic detection is still feasible.

    AB - Automatically generated fake restaurant reviews are a threat to online review systems. Recent research has shown that users have difficulties in detecting machine-generated fake reviews hiding among real restaurant reviews. The method used in this work (char-LSTM) has one drawback: it has difficulties staying in context, i.e. when it generates a review for specific target entity, the resulting review may contain phrases that are unrelated to the target, thus increasing its detectability. In this work, we present and evaluate a more sophisticated technique based on neural machine translation (NMT) with which we can generate reviews that stay on-topic. We test multiple variants of our technique using native English speakers on Amazon Mechanical Turk. We demonstrate that reviews generated by the best variant have almost optimal undetectability (class-averaged F-score 47%). We conduct a user study with experienced users and show that our method evades detection more frequently compared to the state-of-the-art (average evasion 3.2/4 vs 1.5/4) with statistical significance, at level α1% (Sect. 4.3). We develop very effective detection tools and reach average F-score of 97% in classifying these. Although fake reviews are very effective in fooling people, effective automatic detection is still feasible.

    UR - http://www.scopus.com/inward/record.url?scp=85052238662&partnerID=8YFLogxK

    UR - http://www.scopus.com/inward/citedby.url?scp=85052238662&partnerID=8YFLogxK

    U2 - 10.1007/978-3-319-99073-6_7

    DO - 10.1007/978-3-319-99073-6_7

    M3 - Conference contribution

    AN - SCOPUS:85052238662

    SN - 9783319990729

    T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

    SP - 132

    EP - 151

    BT - Computer Security - 23rd European Symposium on Research in Computer Security, ESORICS 2018, Proceedings

    A2 - Lopez, Javier

    A2 - Zhou, Jianying

    A2 - Soriano, Miguel

    PB - Springer-Verlag

    ER -