Users of existing Web information retrieval systems need to input proper keywords through which the correct retrieval data can be found. It is, however, too difficult for children to identify appropriate keywords due to their lack of ability in generalizing concepts describing what they would like to search for: Therefore, a great deal of irrelevant data is included in the results. In the paper, we propose a "Kid's Lifestyle Ontology" so that the system can infer what they would like to search for to resolve ambiguities. The retrieval system can identify appropriate keywords that have strong relationships with concepts common to their lifestyles, and then the related search keywords can be ranked above other keywords. This is based on the episodic memories of children that help them discover other keywords. As a result, children can rejne their queries gradually. In the paper, we also propose a method of constructing the ontology incrementally and recursively.