{"created":"2025-01-19T01:45:26.263962+00:00","metadata":{"_oai":{"id":"oai:ipsj.ixsq.nii.ac.jp:00240952","sets":["6164:6165:6462:11854"]},"path":["11854"],"owner":"11","recid":"240952","title":["大規模言語モデルを用いた悪性JavaScriptの検知手法の提案"],"pubdate":{"attribute_name":"PubDate","attribute_value":"2024-10-15"},"_buckets":{"deposit":"502070ee-ba18-4440-a793-48df369f6e18"},"_deposit":{"id":"240952","pid":{"type":"depid","value":"240952","revision_id":0},"owners":[11],"status":"published","created_by":11},"item_title":"大規模言語モデルを用いた悪性JavaScriptの検知手法の提案","author_link":["662409","662410","662411","662412"],"item_titles":{"attribute_name":"タイトル","attribute_value_mlt":[{"subitem_title":"大規模言語モデルを用いた悪性JavaScriptの検知手法の提案","subitem_title_language":"ja"},{"subitem_title":"Proposal of a method to detect malicious JavaScript  code using Large Language Models","subitem_title_language":"en"}]},"item_keyword":{"attribute_name":"キーワード","attribute_value_mlt":[{"subitem_subject":"大規模言語モデル，Transformer，悪性JavaScript，自然言語処理技術","subitem_subject_scheme":"Other"}]},"item_type_id":"18","publish_date":"2024-10-15","item_language":{"attribute_name":"言語","attribute_value_mlt":[{"subitem_language":"jpn"}]},"item_18_text_3":{"attribute_name":"著者所属","attribute_value_mlt":[{"subitem_text_value":"防衛大学校研究科"},{"subitem_text_value":"防衛大学校研究科"}]},"item_18_text_4":{"attribute_name":"著者所属(英)","attribute_value_mlt":[{"subitem_text_value":"National Defense Academy of Japan","subitem_text_language":"en"},{"subitem_text_value":"National Defense Academy of Japan","subitem_text_language":"en"}]},"item_publisher":{"attribute_name":"出版者","attribute_value_mlt":[{"subitem_publisher":"情報処理学会","subitem_publisher_language":"ja"}]},"publish_status":"0","weko_shared_id":-1,"item_file_price":{"attribute_name":"Billing file","attribute_type":"file","attribute_value_mlt":[{"url":{"url":"https://ipsj.ixsq.nii.ac.jp/record/240952/files/IPSJ-CSS2024206.pdf","label":"IPSJ-CSS2024206.pdf"},"date":[{"dateType":"Available","dateValue":"2026-10-15"}],"format":"application/pdf","billing":["billing_file"],"filename":"IPSJ-CSS2024206.pdf","filesize":[{"value":"2.5 MB"}],"mimetype":"application/pdf","priceinfo":[{"tax":["include_tax"],"price":"660","billingrole":"5"},{"tax":["include_tax"],"price":"330","billingrole":"6"},{"tax":["include_tax"],"price":"0","billingrole":"30"},{"tax":["include_tax"],"price":"0","billingrole":"46"},{"tax":["include_tax"],"price":"0","billingrole":"44"}],"accessrole":"open_date","version_id":"97b2e876-4f61-4bf2-9d11-62a654802b57","displaytype":"detail","licensetype":"license_note","license_note":"Copyright (c) 2024 by the Information Processing Society of Japan"}]},"item_18_creator_5":{"attribute_name":"著者名","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"木下, 慧一"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"三村, 守"}],"nameIdentifiers":[{}]}]},"item_18_creator_6":{"attribute_name":"著者名(英)","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"Keiichi, Kinoshita","creatorNameLang":"en"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"Mamoru, Mimura","creatorNameLang":"en"}],"nameIdentifiers":[{}]}]},"item_resource_type":{"attribute_name":"資源タイプ","attribute_value_mlt":[{"resourceuri":"http://purl.org/coar/resource_type/c_5794","resourcetype":"conference paper"}]},"item_18_description_7":{"attribute_name":"論文抄録","attribute_value_mlt":[{"subitem_description":"悪性JavaScriptを用いたサイバー攻撃は依然として脅威である．これに対し，悪性JavaScriptの言語的特徴を抽出し，自然言語処理技術および機械学習モデルを用いて悪性JavaScriptを高速に検知する手法が提案されている．これらの手法ではJavaScriptを実行しないため動作は高速であるが，未知の悪性JavaScriptの検知率は低いという課題がある．この課題を解決するため，大規模言語モデルを用いた悪性JavaScriptの検知手法を提案する．著者らが知り得る限り，大規模言語モデルを用いた言語的特徴に基づく悪性JavaScriptの検知手法は存在しない．本研究では，Meta社の大規模言語モデルであるLlama2およびCode Llamaを用いて提案手法を実装し，実環境を模擬した不均衡データセットを用いて精度を評価した．その結果，既存手法と比較して未知の悪性JavaScriptの検知率が向上することを確認した．また，Code Llamaは悪性JavaScriptの検知に有効であることを確認した．","subitem_description_type":"Other"}]},"item_18_description_8":{"attribute_name":"論文抄録(英)","attribute_value_mlt":[{"subitem_description":"Cyber attacks using malicious JavaScript code remain a significant threat. In response to this, methods have been proposed to rapidly detect malicious JavaScript code by extracting linguistic features with natural language processing techniques and machine learning models. These methods are fast because they do not execute JavaScript code. However, they have the drawback detection rates for unknown malicious JavaScript code. To address this issue, we propose a method for detecting malicious JavaScript code using large language models. To the best of our knowledge, there is no existing method that uses large language models to detect malicious JavaScript code based on linguistic features. In this study, we implement the proposed method using Meta's large language models, Llama2 and Code Llama, and evaluate the accuracy using an imbalanced dataset that simulate real-world environments. As a result, we confirmed that the detection rate of unknown malicious JavaScript code improved compared to existing methods. Furthermore, we verified that Code Llama is effective in detecting malicious JavaScript code.","subitem_description_type":"Other"}]},"item_18_biblio_info_10":{"attribute_name":"書誌情報","attribute_value_mlt":[{"bibliographicPageEnd":"1560","bibliographic_titles":[{"bibliographic_title":"コンピュータセキュリティシンポジウム2024論文集"}],"bibliographicPageStart":"1553","bibliographicIssueDates":{"bibliographicIssueDate":"2024-10-15","bibliographicIssueDateType":"Issued"}}]},"relation_version_is_last":true,"weko_creator_id":"11"},"id":240952,"updated":"2025-03-06T05:58:21.519834+00:00","links":{}}