東京大学松尾研究室×BAI×リクルートマーケティングパートナーズ 共同研究。学習者の「解けない問題」の予測に成功「つまづき予防」の実現へ:University of Tokyo Matsuo laboratory × BAI × recruit Marketing Partners joint research: Successful prediction of unsolvable problems of learners. Realization of preventing stumbling blocks.

2016. 9. 15

株式会社リクルートマーケティングパートナーズ(本社:東京都中央区、代表取締役社長:山口 文洋)が運営する、リクルート次世代教育研究院(院長:小宮山利恵子)は、東京大学 松尾研究室及び株式会社IGPIビジネスアナリティクス&インテリジェンス(BAI)と実施している「アダプティブ・ラーニング実現に向けた人工知能解析」共同研究の進捗をご報告致します。

Progress report of the joint research program “Artificial Intelligence Analysis for Adaptive Learning” between Recruit Institute of Next Generation Eduction, operated by Recruit Marketing Partners, Matsuo Lab of Tokyo University, and IGPO Business Analytics and Intelligence (BAI). * Recruit Marketing Partners, Chuo-ku Tokyo, CEO Fumihiro YAMAGUCHI * Recruit Institute of Next Generation Education, Director Rieko KOMIYAMA

Analysis of 4th grade to 9th grade Mathematics

教育×人工知能で昨今話題となっている「アダプティブ・ラーニング(学習プロセスの最適化)」において、リクルートマーケティングパートナーズは人工知能研究第一人者である東京大学 松尾研究室及び株式会社IGPIビジネスアナリティクス&インテリジェンス(BAI)と共同研究を行っておりますが、この度『スタディサプリ』における学習者の「解けない問題」を最大30%予測することに成功しました。

Joint research in AI driven adaptive learning between Recruit Marketing Partners and leading AI researchers at the Matsuo Lab of Tokyo University and IGPI Inc., BAI have successfully predicted 30% of “unsolvable problems” encountered by learners using Study Sapuri app.


Deep machine learning analysis was utilized with Study Sapuri learner behavior logs of video viewing and problem solving to identify 30% of questions identified as unsolvable by 5th grade mathematics learners with 90% precision.



The analysis targets students from 4th grade to 9th grade mathematics, with a cut off of 80% precision. 30.4% of the questions that are unsolvable by 5th grade learners were predicted with 90% precision, and 27.1% of unsolvable questions by 6th grade learners were predicted with 88% precision.


Analysis utilized deep knowledge tracing in deep learning, against Study Sapuri learner logs which include video viewing and problem solving data.


We believe predicting unsolvable problems leads to higher efficiency in learning process and time use, in addition to maintaining learner motivation and efficient review of content.


We will continue to research higher learning effectiveness and optimization in conjunction with Associate Professor Yutaka MATSUO and the Matsuo Lab.

Accuracy and coverage of per school year



*1 Precision:
The percentage of unsolvable problems calculated by data analysis to actual unsolvable problems. The collected data is separated into model generation data, and prediction data. The model generation data is used to conduct data analysis and predict the unsolvable problems contained in the prediction data.


*2 Coverage:
The ratio of unsolvable problems calculated by data analysis being an actual unsolvable problem. If there are 10 unsolvable problems and 3 questions were predicted by data analysis, the coverage would be 30%.

From Joint Researchers

東京大学 松尾豊特任准教授のコメント:Yutaka MATSUO, Associate Professor at Tokyo University


Predicting unsolvable problems is an important fundamental technology that will allow us to effectively suggest review and prevent stumbling blocks in learner progress. We are able to provide this level of precision and coverage with the combination of the massive volume of data from Study Sapuri, and AI technology.

松尾豊:東京大学で、人工知能(推論、機械学習、ディープラーニング)、自然言語処理、社会ネットワーク分析、ソーシャルメディア、ウェブマイニング、ビジネスモデルの研究に従事。国内では、人工知能学会を中心に、国際的には,WWW国際会議(International World Wide Web Conference),米国人工知能学会等(AAAI)を中心に論文を発表。WWWではウェブマイニング部門のトラックチェアを務める。現在は、ディープラーニングの研究に注力する一方、国内の企業と,データ分析を中心とする共同研究を行う。東京大学大学院工学系研究科 総合研究機構(若手育成プログラム)/知の構造化センター/技術経営戦略学専攻 准教授を経て、東京大学大学院工学系研究科 技術経営戦略学専攻 消費インテリジェンス寄付講座 代表、シンガポール国立大学(NUS) 客員准教授

Yutaka MATSUO: Researcher at the University of Tokyo, specializing in the study of artificial intelligence (reasoning, machine learning, deep learning), natural language processing, social network analysis, social media, web mining, and business models. In Japan, Professor MATSUO primarily works with the Japanese Society for Artificial Intelligence. Internationally he publishes at the WWW International Conference (International World Wide Web Conference), the United States Society for Artificial Intelligence (AAAI), among other forums.Professor MATSUO serves as track chair of the web mining track with WWW. He currently focuses on deep learning research, and conducts joint data analysis research with Japanese companies. After serving with the University of Tokyo, Graduate School of Engineering, system research department Research Organization (Youth Development Program) / Center of knowledge structure / Associate Professor of technology management strategy, Professor MATSUO is now Endowment Lecture Representative of University of Tokyo, Graduate School of Engineering system research department Technology Management strategy consumption intelligence, and Visiting Associate Professor at the National University of Singapore (NUS)

東京大学 松尾研究室 那須野薫さんのコメント:Kaworu NASUNO, Tokyo University Matsuo Lab


First, we developed a method to predict unsolvable problems (or stumbling blocks for learner progress). In the future, we will work to expand this to cover additional subjects, and develop a system to propose optimal learning paths and schedules based on the stumbling block prediction.

■本研究内容:Research Content
・対象(Target):スタディサプリ小学講座・中学講座(Study Sapuri elementary and middle school courses)
・学年(Grade Level):小学4年〜中学3年(4th to 9th grade )
・ログ数(The number of log):約3600000(approximately 3600000)
・解析データ期間(Data Collection Period):2015年4月~2016年3月(April 2015 to March 2016)

■共同研究概要:Joint Research Overview
・目的(Purpose):アダプティブラーニング実現に向けた人工知能およびビッグデータ解析研究(Artificial intelligence and big data analysis research towards implementation of adaptive learning)
・開始時期(Research Initiated):2014年4月1日(April 1st, 2014)


For further information, please contact