Eisaku Yoshikawa received the Best Interactive Award at the 15th Forum on Data Engineering and Information Management (DEIM 2023) for his presentation. This year, three presentations were selected for the top award out of 342 poster presentations.
His research focuses on exclusion queries for image searches. Exclusion queries are widely used in general web search engines. For example, if someone wants to search for information about the animal “jaguar,” but searching for “jaguar” returns mostly results about Jaguar cars, they can refine the search by using “jaguar -car”, which excludes pages related to cars.
Similarly, keyword-based image search is commonly used in web search engines, utilizing keywords that appear near an image to retrieve relevant images. However, applying this strategy of exclusion queries to the keyword-based image retrieval does not work well. Simply excluding images based on the presence of negative keywords near them often fails to remove unwanted images effectively.
For instance, if someone searches for images of a criminal suspect who went by the name “Luffy” using “Luffy -One_Piece”, many of the top results may still include images of Luffy, the character from the One Piece manga and anime.
To address this issue, he developed a method that, when given a search query in the form of “A -B”, removes images from the “A” search results that are visually similar to images retrieved in an “A B” search. This ensures more effective exclusion of images related to B.
By the way, the figure of Luffy from One Piece visible on the table in the photo.