{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/textbugger-generating-adversarial-text","title":"TextBugger: Generating Adversarial Text Against Real-world Applications","arxiv_id":"1812.05271","date":"2018-12-13","proceeding":null,"authors":["Jinfeng Li","Shouling Ji","Tianyu Du","Bo Li","Ting Wang"],"abstract":"Deep Learning-based Text Understanding (DLTU) is the backbone technique\nbehind various applications, including question answering, machine translation,\nand text classification. Despite its tremendous popularity, the security\nvulnerabilities of DLTU are still largely unknown, which is highly concerning\ngiven its increasing use in security-sensitive applications such as sentiment\nanalysis and toxic content detection. In this paper, we show that DLTU is\ninherently vulnerable to adversarial text attacks, in which maliciously crafted\ntexts trigger target DLTU systems and services to misbehave. Specifically, we\npresent TextBugger, a general attack framework for generating adversarial\ntexts. In contrast to prior works, TextBugger differs in significant ways: (i)\neffective -- it outperforms state-of-the-art attacks in terms of attack success\nrate; (ii) evasive -- it preserves the utility of benign text, with 94.9\\% of\nthe adversarial text correctly recognized by human readers; and (iii) efficient\n-- it generates adversarial text with computational complexity sub-linear to\nthe text length. We empirically evaluate TextBugger on a set of real-world DLTU\nsystems and services used for sentiment analysis and toxic content detection,\ndemonstrating its effectiveness, evasiveness, and efficiency. For instance,\nTextBugger achieves 100\\% success rate on the IMDB dataset based on Amazon AWS\nComprehend within 4.61 seconds and preserves 97\\% semantic similarity. We\nfurther discuss possible defense mechanisms to mitigate such attack and the\nadversary's potential countermeasures, which leads to promising directions for\nfurther research.","url_abs":"http://arxiv.org/abs/1812.05271v1","url_pdf":"http://arxiv.org/pdf/1812.05271v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"textbugger-generating-adversarial-text","repo_url":"https://github.com/CatherineWong/dancin_seq2seq","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"pytorch","reach":{"status":"ok"}}],"tasks":[{"task_slug":"adversarial-text","task_name":"Adversarial Text"},{"task_slug":"machine-translation","task_name":"Machine Translation"},{"task_slug":"question-answering","task_name":"Question Answering"},{"task_slug":"semantic-similarity","task_name":"Semantic Similarity"},{"task_slug":"semantic-textual-similarity","task_name":"Semantic Textual Similarity"},{"task_slug":"sentiment-analysis","task_name":"Sentiment Analysis"},{"task_slug":"text-classification","task_name":"Text Classification"},{"task_slug":"text-classification-1","task_name":"text-classification"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1812.05271","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}