An aproach to high dimensional regression problems on small text data using topic models

2016 Fiscal Year Final Research Report

• Project/Area Number: 15K12149
• Research Category: Grant-in-Aid for Challenging Exploratory Research
• Allocation Type: Multi-year Fund
• Research Field: Web informatics, Service informatics
• Research Institution: University of Tsukuba
• Principal Investigator: YAMAMOTO Mikio 筑波大学, システム情報系, 教授 (40210562)
• Research Collaborator: TSUNODA Takaaki ; YAMAGUCHI Taichi
• Project Period (FY): 2015-04-01 – 2017-03-31
• Keywords: トピックモデル / 縮小推定 / 検索行動量 / 状態空間モデル

Outline of Final Research Achievements

In this research, we investigated (1) high dimensional regression problems using topic models for small text data and (2) prediction problems of car sales using state space models with the search behavior of users on the web. The achievements are the followings.
(1) We showed that various shrinkage estimation methods such as ridge and lasso regressions are effective in order to improve supervised topic models for high dimensional and small text data.
(2) We showed that the search behavior volume data can be used for increasing the accuracy of car sales prediction using state space models.

Free Research Field

情報工学