Active Learning with Partially Annotated Sequence

• Item type: SIG Technical Reports(1)
• 公開日: 2010-09-09
• タイトル: Active Learning with Partially Annotated Sequence
• タイトル / 言語: en
• タイトル: Active Learning with Partially Annotated Sequence
• 言語: eng
• キーワード / 主題Scheme: Other
• キーワード / 主題: 学習・応用
• 資源タイプ / 資源タイプ識別子: http://purl.org/coar/resource_type/c_18gh
• 資源タイプ: technical report
• 著者所属: Department of Computational Intelligence and Systems Science, Tokyo Institute of Technology
• 著者所属: Precision and Intelligence Laboratory, Tokyo Institute of Technology
• 著者所属: Precision and Intelligence Laboratory, Tokyo Institute of Technology
• 著者所属(英): en
• 著者所属(英): Department of Computational Intelligence and Systems Science, Tokyo Institute of Technology
• 著者所属(英): en
• 著者所属(英): Precision and Intelligence Laboratory, Tokyo Institute of Technology
• 著者所属(英): en
• 著者所属(英): Precision and Intelligence Laboratory, Tokyo Institute of Technology
• 著者名: Dittaya, Wanvarie; Hiroya, Takamura; Manabu, Okumura
• 著者名(英): Dittaya, Wanvarie; Hiroya, Takamura; Manabu, Okumura
• 論文抄録 / 内容記述タイプ: Other
• 論文抄録 / 内容記述: We propose an active learning framework which requires human annotation only in the ambiguous parts of the sequence. In each iteration of active learning, a set of tokens from the ambiguous parts are manually labeled while the other tokens are left unannotated. Our proposed method is superior to the method where unambiguous tokens are automatically labeled. We evaluate our proposed framework on chunking and named entity recognition data provided by CoNLL. Experimental results show that our proposed framework outperforms the previous work using automatically labeled tokens, and almost reaches the supervised F1 with 6.37% and 8.59% of tokens being manually labeled, respectively.
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: We propose an active learning framework which requires human annotation only in the ambiguous parts of the sequence. In each iteration of active learning, a set of tokens from the ambiguous parts are manually labeled while the other tokens are left unannotated. Our proposed method is superior to the method where unambiguous tokens are automatically labeled. We evaluate our proposed framework on chunking and named entity recognition data provided by CoNLL. Experimental results show that our proposed framework outperforms the previous work using automatically labeled tokens, and almost reaches the supervised F1 with 6.37% and 8.59% of tokens being manually labeled, respectively.
• 書誌レコードID / 収録物識別子タイプ: NCID
• 書誌レコードID / 収録物識別子: AN10115061
• 書誌情報: 研究報告自然言語処理（NL）; 巻 2010-NL-198, 号 4, p. 1-7, 発行日 2010-09-09
• Notice: SIG Technical Reports are nonrefereed and hence may later appear in any journals, conferences, symposia, etc.
• 出版者 / 言語: ja
• 出版者: 情報処理学会