SAKAI, Tetsuya

写真a

Affiliation

Faculty of Science and Engineering, School of Fundamental Science and Engineering

Job title

Professor

Mail Address

E-mail address

Homepage URL

http://sakailab.com/tetsuya/

Profile

http://sakailab.com/tetsuya/

Concurrent Post 【 display / non-display

  • Faculty of Science and Engineering   Graduate School of Fundamental Science and Engineering

Research Institute 【 display / non-display

  • 2020
    -
    2022

    理工学術院総合研究所   兼任研究員

Degree 【 display / non-display

  • 博士

 

Research Areas 【 display / non-display

  • Human interface and interaction

Research Interests 【 display / non-display

  • information access, information retrieval, natural language processing

Papers 【 display / non-display

  • Randomised vs. Prioritised Pools for Relevance Assessments: Sample Size Considerations

    Sakai. T, Xiao, P

    Proceedings of AIRS 2019    2020

  • Generating Short Product Descriptors based on Very Little Training Data

    Xiao, P, Lee, J.-Y, Tao, S, Hwang, Y.-S, Sakai, T

    Proceedings of AIRS 2019    2020

  • Unsupervised Answer Retrieval with Data Fusion for Community Question Answering

    Kato, S, Shimizu, T, Fujita, S, Sakai, T

    Proceedings of AIRS 2019    2020

  • Towards Automatic Evaluation of Reused Answers in Community Question Answering

    Liu, H.-W, Fujita, S, Sakai, T

    Proceedings of AIRS 2019    2020

  • Arc Loss: Softmax with Additive Angular Margin for Answer Retrieval

    Suzuki, R, Fujita, S, Sakai, T

    Proceedings of AIRS 2019    2020

display all >>

Books and Other Publications 【 display / non-display

  • Proceedings of the Open-Source IR Replicability Challenge (OSIRRC 2019)

    Clancy, R, Ferro, N, Hauff, C, Lin, J, Sakai, T, Wu, Z.-Z

    2019

  • Laboratory Experiments in Information Retrieval: Sample Sizes, Effect Sizes, and Statistical Power

    Sakai, T

    Springer  2018

  • Proceedings of AIRS 2018 (LNCS 11292)

    Tseng, Y.-H, Sakai, T, Jiang, J, Ku, L.-W., Park, D.H., Yeh, J.-F., Yu, L.-C, Lee, L.-H, Chen, Z.-H

    2018

  • Proceedings of SPIRE 2016 (LNCS 9954)

    Inegaga, S, Sadakane, K, Sakai, T

    2016

  • 情報アクセス評価方法論~検索エンジンの進歩のために~,

    酒井哲也

    コロナ社  2015

display all >>

Misc 【 display / non-display

  • "ベイズ統計を用いた文書ファイルの自動分析手法,"

    後藤正幸, 伊藤潤, 石田崇, 酒井哲也

    経営情報学会2003年度秋季全国研究発表大会予稿集,函館   pp.28-31  2003

  • 「インターネットを用いた研究活動支援システム」システム構成

    平澤茂一, 松嶋敏泰, 鴻巣敏之, 酒井哲也, 中澤真, 李相協, 野村亮

    2001PCカンファレンス    2001

Awards 【 display / non-display

  • CSS 2019 Best Paper Award (fifth author)

    2019  

  • ACM Distinguished Member

    2018  

  • WASEDA e-Teaching Award 2018

    2018  

  • ACM Recognition of Service Award (SIGIR'17 Co-chair)

    2017  

  • ACM Senior Member

    2016  

display all >>

Research Projects 【 display / non-display

  • ナゲットに基づくタスク指向対話の自動評価に関する研究

    Project Year :

    2017.04
    -
    2021.03
     

     View Summary

    コンペティション型国際会議NTCIR-14にてShort Text Conversation (STC-3) タスクをスケジュール通りに運営し、早稲田大学酒井研究室を含む12の研究機関から結果を提出してもらうことができた。このタスクは、顧客・ヘルプデスク間の対話の品質を推定するものであり、この技術は将来的に対話システムの応答戦略に応用可能である。タスクの評価方法については情報検索会議の最高峰SIGIRにて発表を行い、データセットに関してはJournal of Information Processingにてまとめた。後者はWebDB Forum 2018にてbest paper runner-upに選出された。<BR>・Zeng, Z., Luo, C., Shang, L., Li, H., and Sakai, T.: Towards Automatic Evaluation of Customer-Helpdesk Dialogues, Journal of Information Processing, Volume 26, pp.768-778, 査読あり, 2018. WebDB Forum 2018 Best Paper Runner-up<BR>・Sakai, T.: Comparing Two Binned Probability Distributions for Information Access Evaluation, Proceedings of ACM SIGIR 2018, pp.1073-1076, 査読あり, 2018.以下のスケジュールに沿ってタスク運営を進めることができた。4月 データのクローリング+アノテーションツールの開発、5-8月 データのアノテーション、9月 学習用データ公開、11月 評価用データ公開・結果提出締切、2月 タスクオーバービュー論文暫定版公開、3月 タスク参加者論文暫定版投稿2019年度の計画は以下の通りである。・NTCIR-14にてタスク運営者およびタスク参加者としての研究成果を発表・対話データセットDCH-1の中英翻訳を進め、より広くの対話研究者が使えるようにする・NTCIR-15における対話タスクの設計と提案、推

  • Exploratory Search Considering the User's Situation

    Project Year :

    2016.04
    -
    2020.03
     

Presentations 【 display / non-display

  • 擬似アノテーションにもとづく日本語ツイートの極性判定

    小橋賢介, 酒井哲也

    DEIM 2019 

    Presentation date: 2019

  • FigureQAタスクにおける抽象画像を考慮したアプローチ

    坂本凜, 酒井哲也

    DEIM 2019 

    Presentation date: 2019

  • Convolutional Neural Networkを用いたFake News Challengeの検討

    雨宮佑基, 酒井哲也

    DEIM 2019 

    Presentation date: 2019

  • 音声ユーザインタフェースにおける処理エラーによるユーザフラストレーションに関する調査

    呉越思瑶, 酒井哲也

    DEIM 2019 

    Presentation date: 2019

  • Query-Focused Extractive Summarization based on Deep Learning: Comparison of Similarity Measures for Pseudo Ground Truth Generation

    Yuliska, Tetsuya Sakai

    DEIM 2019 

    Presentation date: 2019

display all >>

Specific Research 【 display / non-display

  • ベイズ統計に基づく情報アクセス評価体系の構築

    2017  

     View Summary

    I published the following full paper at SIGIR 2017, the top conference in information retrieval.The following is the abstract:Using classical statistical signifi€cance tests, researchers can onlydiscuss P(D+|H), the probability of observing the data D at hand orsomething more extreme, under the assumption that the hypothesisH is true (i.e., the p-value). But what we usually want is P(D+|H),the probability that a hypothesis is true, given the data. If we useBayesian statistics with state-of-the-art Markov Chain Monte Carlo(MCMC) methods for obtaining posterior distributions, this is nolonger a problem. Th‘at is, instead of the classical p-values and 95%confi€dence intervals, which are oft‰en misinterpreted respectivelyas “probability that the hypothesis is (in)correct” and “probabilitythat the true parameter value drops within the interval is 95%,” wecan easily obtain P(H|D) and credible intervals which representexactly the above. Moreover, with Bayesian tests, we can easilyhandle virtually any hypothesis, not just “equality of means,” andobtain an Expected A Posteriori (EAP) value of any statistic thatwe are interested in. We provide simple tools to encourage theIR community to take up paired and unpaired Bayesian tests forcomparing two systems. Using a variety of TREC and NTCIR data,we compare P(H|D) with p-values, credible intervals with con€fidence intervals, and Bayesian EAP eff‚ect sizes with classical ones.Our results show that (a) p-values and confi€dence intervals canrespectively be regarded as approximations of what we really want,namely, P(H|D) and credible intervals; and (b) sample eff‚ect sizesfrom classical signifi€cance tests can diff‚er considerably from theBayesian EAP eff‚ect sizes, which suggests that the former can bepoor estimates of population e‚ffect sizes. For both paired and unpairedtests, we propose that the IR community report the EAP, thecredible interval, and the probability of hypothesis being true, notonly for the raw diff‚erence in means but also for the eff‚ect size interms of Glass’s delta.

  • 統計的手法を用いた情報検索テストコレクション横断評価および情報検索論文の評価

    2016  

     View Summary

    I published five international conference papers (SIGIR, SIGIR, SIGIR(short), ICTIR, AIRS),two international workshop papers (EVIA, EVIA), and a workshop report (SIGIR Forum).Moreover, I gave a tutorial at an international conference (ICTIR) and a keynote at a Japanese symposium (IPSJ SIGNL) on this topic.

  • 「寡黙なユーザ」のための情報検索技術に関する研究

    2015  

     View Summary

    We published one international journal paper, one international conference paper, one evaluation conference overview (TREC), and two unrefereed domestic papers.

  • 情報アクセス評価基盤の体系化および評価

    2015  

     View Summary

    We published one book, one international journal paper, one international conference paper, one domestic IPSJ workshop paper and organised an international workshop.

  • サーチエンジン評価指標の体系化と有効性実証

    2014  

     View Summary

    We published three refereed papers (two forinternational conferences and one for a domestic conference) on how todetermine the topic set size of a test collection.

display all >>

 

Syllabus 【 display / non-display

display all >>