{"updated":"2025-01-19T07:36:15.249805+00:00","metadata":{"_oai":{"id":"oai:ipsj.ixsq.nii.ac.jp:00241611","sets":["1164:4179:11560:11869"]},"path":["11869"],"owner":"44499","recid":"241611","title":["長文要約タスクにおける複数学習条件の効果検証"],"pubdate":{"attribute_name":"公開日","attribute_value":"2024-12-05"},"_buckets":{"deposit":"2bd6121a-be02-4d47-9691-2a0169040f71"},"_deposit":{"id":"241611","pid":{"type":"depid","value":"241611","revision_id":0},"owners":[44499],"status":"published","created_by":44499},"item_title":"長文要約タスクにおける複数学習条件の効果検証","author_link":["665473","665472","665471"],"item_titles":{"attribute_name":"タイトル","attribute_value_mlt":[{"subitem_title":"長文要約タスクにおける複数学習条件の効果検証"},{"subitem_title":"Effect Verification of Multiple Learning Conditions on Long-text Summarization Tasks","subitem_title_language":"en"}]},"item_keyword":{"attribute_name":"キーワード","attribute_value_mlt":[{"subitem_subject":"要約","subitem_subject_scheme":"Other"}]},"item_type_id":"4","publish_date":"2024-12-05","item_4_text_3":{"attribute_name":"著者所属","attribute_value_mlt":[{"subitem_text_value":"コニカミノルタ株式会社／現在，日本電信電話株式会社人間情報研究所"},{"subitem_text_value":"コニカミノルタ株式会社"},{"subitem_text_value":"コニカミノルタ株式会社"}]},"item_language":{"attribute_name":"言語","attribute_value_mlt":[{"subitem_language":"jpn"}]},"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/241611/files/IPSJ-NL24262036.pdf","label":"IPSJ-NL24262036.pdf"},"date":[{"dateType":"Available","dateValue":"2026-12-05"}],"format":"application/pdf","billing":["billing_file"],"filename":"IPSJ-NL24262036.pdf","filesize":[{"value":"338.4 kB"}],"mimetype":"application/pdf","priceinfo":[{"tax":["include_tax"],"price":"660","billingrole":"5"},{"tax":["include_tax"],"price":"330","billingrole":"6"},{"tax":["include_tax"],"price":"0","billingrole":"23"},{"tax":["include_tax"],"price":"0","billingrole":"44"}],"accessrole":"open_date","version_id":"7fa5764c-a68d-414a-99db-7fe508e5bcab","displaytype":"detail","licensetype":"license_note","license_note":"Copyright (c) 2024 by the Information Processing Society of Japan"}]},"item_4_creator_5":{"attribute_name":"著者名","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"樋本, 一晴"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"池田, 大志"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"寺中, 駿人"}],"nameIdentifiers":[{}]}]},"item_4_source_id_9":{"attribute_name":"書誌レコードID","attribute_value_mlt":[{"subitem_source_identifier":"AN10115061","subitem_source_identifier_type":"NCID"}]},"item_4_textarea_12":{"attribute_name":"Notice","attribute_value_mlt":[{"subitem_textarea_value":"SIG Technical Reports are nonrefereed and hence may later appear in any journals, conferences, symposia, etc."}]},"item_resource_type":{"attribute_name":"資源タイプ","attribute_value_mlt":[{"resourceuri":"http://purl.org/coar/resource_type/c_18gh","resourcetype":"technical report"}]},"item_4_source_id_11":{"attribute_name":"ISSN","attribute_value_mlt":[{"subitem_source_identifier":"2188-8779","subitem_source_identifier_type":"ISSN"}]},"item_4_description_7":{"attribute_name":"論文抄録","attribute_value_mlt":[{"subitem_description":"本稿は，長文要約タスクにおける，大規模言語モデル (Large Language Model: LLM) への複数の学習条件の効果を検証した結果を報告する．長文要約タスクの実現手段としては，LLM の学習以外にも，Instruct チューニング済み LLM を活用した zero-shot や few-shot 推論が考えられる．長文要約タスクを実適用する際，アノテーション済みデータセットを構築できる場合は，LLM に学習を加えることで，タスク精度の向上に取り組むことが一般的である．ただし，アノテーションデータの構築は多くの工数がかかることが知られている．そこで，本研究では，長文要約タスクの実現に対し，必要なアノテーションデータ数に対する精度比較を行った．実験では，モデルパラメータ数が 7B の Decoder モデルに対するファインチューニングと LoRA チューニング，軽量な Encoder-Decoder モデルと，Encoder モデルに対するファインチューニングを行い，それぞれ学習データ数に対する要約精度の比較を行った．その結果，(1) モデルパラメータ数が 7B の LLM に対して，ファインチューニングと LoRA チューニングの精度差はあまり見られない．(2) 学習データ数が少ない場合は，モデルパラメータ数が 7B の LLM に対する LoRA チューニングが高精度だが，学習データ数が増加するにつれ，軽量な Encoder-Decoder モデルに対するファインチューニングの方が高精度になる．の大きく二つの傾向がわかり，長文要約タスクの実適用に対し，アノテーションデータ数に応じた現実的な解法の方針を示した．","subitem_description_type":"Other"}]},"item_4_biblio_info_10":{"attribute_name":"書誌情報","attribute_value_mlt":[{"bibliographicPageEnd":"8","bibliographic_titles":[{"bibliographic_title":"研究報告自然言語処理（NL）"}],"bibliographicPageStart":"1","bibliographicIssueDates":{"bibliographicIssueDate":"2024-12-05","bibliographicIssueDateType":"Issued"},"bibliographicIssueNumber":"36","bibliographicVolumeNumber":"2024-NL-262"}]},"relation_version_is_last":true,"weko_creator_id":"44499"},"created":"2025-01-19T01:46:19.085497+00:00","id":241611,"links":{}}