How to Search for Information Quickly and Accurately: Exploring the Theory and Technology of Information Retrieval Systems

Although convenient information retrieval systems like Google are now widespread, my research focuses on the theory and technology for advancing these systems, including search engines. I conduct research that aims to improve search performance using statistical methods. I also study methodologies for search experiments and clustering techniques for automatically sorting large volumes of documents.

One area where I am beginning to see significant results is in the research of cross-lingual information retrieval systems. A cross-lingual information retrieval system allows for searching across different languages—for instance, entering a query in Japanese to search an English database and display the results. Since automatic translation does not yet yield satisfactory results, I devised a system that uses sophisticated statistical processing to rank the desired answers higher. I have developed several methods, and their performance verification is nearly complete.

Currently, I am focusing my efforts on the study of document clustering. Document clustering is a method for automatically classifying a large, disorganized collection of documents by grouping similar ones together. For example, if a month's worth of news from a news agency were automatically classified and labeled into categories such as articles about cars or sports, the information would become much more accessible.

Classification requires not only examining each document individually but also checking pairs of documents to determine their similarity. Consequently, with a large number of documents, the computational load becomes enormous, which can cause the process to stall. While this has been a research topic for a long time, I am building on the existing work and tackling this research to see if I can make a proposal from a new perspective.

Document clustering also helps improve search performance and provides search assistance. Clustering search engines have recently emerged that automatically sort and label hundreds of search results. Applications in areas such as this are highly anticipated.

This research is rewarding because its results have practical applications. I enjoy programming, and I also appreciate that experiments yield clear, definitive results. Much like in the natural sciences, I like being able to make steady progress by building upon the accumulation of past research findings and knowledge. I also frequently conduct research in a workshop format, sharing themes and solving problems with researchers from around the world. I find this style enjoyable as well.