| Whether we're corporate employees or consumers, we do not always find what we're looking for when searching databases. Search phrases are often at low levels of abstraction (like office chair, as opposed to higher levels of abstraction like chair or furniture). Most user queries are formulated with text in which users generally enter two-word phrases and expect to receive a cluster of results meeting their needs. However, multimedia databases (like ebay.com or amazon.com) offer the opportunity to formulate queries with images that, in turn, represent clusters of search results. Though much is known about categorization at high levels of abstraction, very little research has been done in the area of using images to represent categories at low levels of abstraction.;Following categorization of 60 images of objects (office chairs ) through a multiple pile-sort technique, visual query interfaces utilizing hierarchies of typical and atypical images to represent clusters of the office chairs at low levels of abstraction were created. Using a search task, this dissertation examines the effect of typical images on search time and argues that as similarity increases, the effect of typicality on search efficacy is lessened. A 2 (typicality: typical vs. atypical) x 2 (similarity: similar vs. dissimilar) shows that typical images facilitate faster searching compared to atypical images, but that this effect is reduced, and even reversed, when these objects are very similar to each other. Implications for information systems, such as search engines, and marketing are discussed. |
