2023 Fiscal Year Final Research Report

A scalable privacy-preserving information retrieval system based on federated optimization, on-device intelligence and semantic matching

Research Project

PDF

Project/Area Number	19H04215
Research Category	Grant-in-Aid for Scientific Research (B)
Allocation Type	Single-year Grants
Section	一般
Review Section	Basic Section 62020:Web informatics and service informatics-related
Research Institution	University of Tsukuba
Principal Investigator	Yu Haitao 筑波大学, 図書館情報メディア系, 准教授 (30751052)
Co-Investigator(Kenkyū-buntansha)	吉川正俊大阪成蹊大学, データサイエンス学部, 教授 (30182736) 康シン徳島大学, 大学院社会産業理工学研究部(理工学域), 助教 (80777350) Adam Jatowt 京都大学, 情報学研究科, 特定准教授 (00415861)
Project Period (FY)	2019-04-01 – 2024-03-31
Keywords	Federated Learning / Large Language Model / Conversational IR / Generative IR / Personalization / On-device Learning / User Modeling / Neural Tree Ensemble
Outline of Final Research Achievements	This project aims to initiate research into privacy-preserving information retrieval (IR). Throughout the lifetime of this project, we made remarkable achievements via the following main aspects: novel ways of combining federated learning and differential privacy for privacy-preserving information access, direct optimization of evaluation metrics, effective integration of LLMs for result re-ranking, incorporating personalized context for conversational information seeking. As a result, we published more than 20 conference papers and 10 journal papers. Moreover, we are also maintaining an open-source project for IR named as PTRanking, which includes many representative ranking methods based on neural networks. Overall, it is reasonable to say that the successful accomplishment of this project will bring new insights into the development of privacy-preserving IR techniques.
Free Research Field	情報検索
Academic Significance and Societal Importance of the Research Achievements	Our research achievements would deepen the understanding of privacy-preserving information seeking that goes beyond information retrieval (IR). By releasing the source codes and collections, we encourage the entire IR community to improve the research of privacy-preserving IR towards new stages.