Hierarchical Classification from Big Data

2015 Fiscal Year Final Research Report

• Project/Area Number: 25330271
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Research Field: Intelligent informatics
• Research Institution: Toyota Technological Institute
• Principal Investigator: Yutaka Sasaki 豊田工業大学, 工学(系)研究科(研究院), 教授 (60395019)
• Project Period (FY): 2013-04-01 – 2016-03-31
• Keywords: 階層的分類 / 機械学習 / LSHTC3 / 分散ベクトル表現 / 文書分類 / Big Data

Outline of Final Research Achievements

We constructed fast and accurate hierarchical classification systems on the basis of the LSHTC3 Wikipedia data, which are huge hierarchical classification datasets. The training time of our system on the LSHTC3 Wikipedia Medium data has been reduced to 30 minutes. Conventional methods for the same data took several hours or even several days. The predictive performance for the test data showed the world highest scores. Moreover, we generated new features based on the distributed embedding vectors which have been created from the original features. Adding the new features further improved the predictive performance over the test data to 44.92%. We made our hierarchical classification system Eze publicly available as open-source software.

Free Research Field

人工知能