{"created":"2025-11-26T01:58:08.696510+00:00","metadata":{"_oai":{"id":"oai:ipsj.ixsq.nii.ac.jp:02005984","sets":["6164:6165:6244:1764117461240"]},"path":["1764117461240"],"owner":"80578","recid":"2005984","title":["機械学習とLLMを用いた多段階フィッシング検知手法の設計と実装"],"pubdate":{"attribute_name":"PubDate","attribute_value":"2025-12-04"},"_buckets":{"deposit":"81540fa8-54a4-4410-a01e-717a6473ae32"},"_deposit":{"id":"2005984","pid":{"type":"depid","value":"2005984","revision_id":0},"owners":[80578],"status":"published","created_by":80578},"item_title":"機械学習とLLMを用いた多段階フィッシング検知手法の設計と実装","author_link":[],"item_titles":{"attribute_name":"タイトル","attribute_value_mlt":[{"subitem_title":"機械学習とLLMを用いた多段階フィッシング検知手法の設計と実装","subitem_title_language":"ja"},{"subitem_title":"Design and Implementation of a Multi-Stage Phishing Detection Method Using Machine Learning and LLMs","subitem_title_language":"en"}]},"item_keyword":{"attribute_name":"キーワード","attribute_value_mlt":[{"subitem_subject":"ネットワークセキュリティ","subitem_subject_scheme":"Other"}]},"item_type_id":"18","publish_date":"2025-12-04","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":"東北大学大学院情報科学研究科"},{"subitem_text_value":"San Francisco Xavier de Chuquisaca University"},{"subitem_text_value":"仙台高等専門学校"},{"subitem_text_value":"東北大学サイバーサイエンスセンター／東北大学大学院情報科学研究科"}]},"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/2005984/files/IPSJ-IOTS2025006.pdf","label":"IPSJ-IOTS2025006.pdf"},"date":[{"dateType":"Available","dateValue":"2027-12-04"}],"format":"application/pdf","billing":["billing_file"],"filename":"IPSJ-IOTS2025006.pdf","filesize":[{"value":"1.8 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":"43"},{"tax":["include_tax"],"price":"0","billingrole":"44"}],"accessrole":"open_date","version_id":"e4bf705f-85d8-4619-a1d4-a570f0e1fc44","displaytype":"detail","licensetype":"license_note","license_note":"Copyright (c) 2025 by the Information Processing Society of Japan"}]},"item_18_creator_5":{"attribute_name":"著者名","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"吉田,蓮"}]},{"creatorNames":[{"creatorName":"ギリエ,ルイス"}]},{"creatorNames":[{"creatorName":"和泉,諭"}]},{"creatorNames":[{"creatorName":"菅沼,拓夫"}]}]},"item_18_creator_6":{"attribute_name":"著者名(英)","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"Ren Yoshida","creatorNameLang":"en"}]},{"creatorNames":[{"creatorName":"Luis Guillen","creatorNameLang":"en"}]},{"creatorNames":[{"creatorName":"Satoru Izumi","creatorNameLang":"en"}]},{"creatorNames":[{"creatorName":"Takuo Suganuma","creatorNameLang":"en"}]}]},"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":"フィッシング攻撃が高度化する中，従来の検知手法は容易に回避される．特に，大規模言語モデル（LLM）の普及により，フィッシング攻撃の作成と展開は容易になっている．しかし，LLMはこれらの攻撃を検知・防御するためにも利用可能である．本論文では，機械学習と複数のLLMを統合し，検知性能と効率性能の向上を目指した多段階フィッシング検知システムを提案する．提案システムは，堅牢かつ効率的なパイプラインを採用しており，マルチモーダル入力に対して初期段階では高速なURL評価を行い，MLによるリスク分析結果に応じて詳細なマルチモーダル評価へと進むことができる．代表的なデータセットおよびサイトを用いた実験結果から，本手法は99.0%の再現率を達成し，既存手法を上回る性能を示した．さらに，従来手法が単純な二値判定にとどまるのに対し，提案システムはコスト効率よく詳細な診断レポートを提供することが可能である．","subitem_description_type":"Other"}]},"item_18_description_8":{"attribute_name":"論文抄録(英)","attribute_value_mlt":[{"subitem_description":"Given the increasing sophistication of phishing attacks, conventional detection methods are easily circumvented. In particular, the proliferation of Large Language Models (LLMs) has facilitated the creation and deployment of these attacks. However, LLMs can also be utilized to detect and defend against them. This paper proposes a multi-stage phishing detection system that integrates machine learning (ML) and multiple LLMs, aiming to improve both detection performance and efficiency. The proposed system employs a robust and efficient pipeline; it initially performs a high-speed URL evaluation on multi-modal inputs, and then proceeds to a detailed multi-modal assessment based on the results of an ML-driven risk analysis. Experimental results using representative datasets and sites demonstrate that our approach achieves 99.0% recall, indicating performance superior to existing methods. Furthermore, while conventional methods are limited to simple binary classification, our proposed system can cost-effectively provide detailed diagnostic reports.","subitem_description_type":"Other"}]},"item_18_biblio_info_10":{"attribute_name":"書誌情報","attribute_value_mlt":[{"bibliographicPageEnd":"48","bibliographic_titles":[{"bibliographic_title":"インターネットと運用技術シンポジウム論文集"}],"bibliographicPageStart":"41","bibliographicIssueDates":{"bibliographicIssueDate":"2025-12-04","bibliographicIssueDateType":"Issued"},"bibliographicVolumeNumber":"2025"}]},"relation_version_is_last":true,"weko_creator_id":"80578"},"id":2005984,"updated":"2025-11-26T07:01:08.702608+00:00","links":{}}