High Precision Information Retrieval and Recommendation based on Copulas

2017 Fiscal Year Final Research Report

• Project/Area Number: 15H02701
• Research Category: Grant-in-Aid for Scientific Research (B)
• Allocation Type: Single-year Grants
• Section: 一般
• Research Field: Multimedia database
• Research Institution: Tokyo Institute of Technology
• Principal Investigator: Miyazaki Jun 東京工業大学, 情報理工学院, 教授 (40293394)
• Co-Investigator(Kenkyū-buntansha): 波多野 賢治 同志社大学, 文化情報学部, 教授 (80314532) ; 中村 匡秀 神戸大学, システム情報学研究科, 准教授 (30324859) ; 欅 惇志 東京工業大学, 情報理工学院, 助教 (00733958)
• Project Period (FY): 2015-04-01 – 2018-03-31
• Keywords: 情報検索 / コピュラ / 情報推薦 / スコア統合

Outline of Final Research Achievements

In this research, we applied copulas which can consider complex dependencies among multiple features to the areas of information retrieval (IR) and recommender systems, and showed a method to design transparent and highly effective IR and recommender systems.
More specifically, we considered a mixture copula model which integrates multiple copulas with a linear combination for building effective IR and recommender systems. To estimate a good mixture copula which affects their effectiveness, we indicated that it is appropriate that a density-based clustering algorithm is applied in the copula estimation phase. In addition, we also developed an efficient top-k algorithm for quickly returning relevant results even if the scoring function is non-linear, such as copulas, and non-monotonic. Moreover, as for recommender systems, we showed that effective recommender systems can also be designed with mixture copulas, when preprocessing feature parameters with a statistical approach.

Free Research Field

データ工学