論文タイトル
Entity Query Feature Expansion using Knowledge Base Links (SIGIR '14)
著者(所属)
Jeffrey Dalton, Laura Dietz, James Allan
(Center for Intelligent Information Retrieval, School of Computer Science, University of Massachusetts Amherst, Amherst, Massachusetts)
次に読むべき論文は?
Schuhmacher, Michael, Laura Dietz, and Simone Paolo Ponzetto. "Ranking entities for web queries through text and knowledge." Proceedings of the 24th ACM international on conference on information and knowledge management. 2015.
論文タイトル Entity Query Feature Expansion using Knowledge Base Links (SIGIR '14)
著者(所属) Jeffrey Dalton, Laura Dietz, James Allan (Center for Intelligent Information Retrieval, School of Computer Science, University of Massachusetts Amherst, Amherst, Massachusetts)
論文PDF/ランディングページへのリンク https://dl.acm.org/doi/10.1145/2600428.2609628
論文まとめ(落合フォーマット準拠)
どんなもの(どんな問題を解きたい)? テキスト内のエンティティとアノテーションされた情報を使ってドキュメント検索をどのように向上させるかという問題に対して取り組んでいる(2014年時点では,open research areaとのこと).
先行研究と比べてどこがすごい? クエリ拡張にエンティティリンキングによって取得できるKBの情報を利用したおそらく最初の研究.
技術や手法のキモはどこ? エンティティに関する様々な特徴量を対数線型モデルに組み込み,情報取得に最適となるように学習を行なった(Learning-to-rank)
どうやって有効だと検証した? ベースライン手法は,Sequencial Dependence Model(SDM),その他2つの手法で比較した. データセットは,Robust04,ClueWeb09,ClueWeb12で,Titleに記載されている単語をクエリとした. MAP, Precision@20, NDCG@20で評価し,有意な差で提案手法がよいと主張しているが,実用的に優位かは微妙な印象.
議論はある? Entity linkingの精度に依存してしまうこと.
次に読むべき論文は? Schuhmacher, Michael, Laura Dietz, and Simone Paolo Ponzetto. "Ranking entities for web queries through text and knowledge." Proceedings of the 24th ACM international on conference on information and knowledge management. 2015.