Vandamme Stijn, De Turck Filip
Department of Information Technology (INTEC), Ghent University-iMinds, Gaston Crommenlaan 8, Bus 201, 9050 Gent, Belgium.
ScientificWorldJournal. 2014 Jan 29;2014:454868. doi: 10.1155/2014/454868. eCollection 2014.
The common user interface for a search engine consists of a text field where the user can enter queries consisting of one or more keywords. Keyword query based search engines work well when the users have a clear vision what they are looking for and are capable of articulating their query using the same terms as indexed. For our multimedia database containing 202,868 items with text descriptions, we supplement such a search engine with a category-based interface whose category structure is tailored to the content of the database. This facilitates browsing and offers the users the possibility to look for named entities, even if they forgot their names. We demonstrate that this approach allows users who fail to recollect the name of named entities to retrieve data with little effort. In all our experiments, it takes 1 query on a category and on average 2.49 clicks, compared to 5.68 queries on the database's traditional text search engine for a 68.3% success probability or 6.01 queries when the user also turns to Google, for a 97.1% success probability.
搜索引擎的通用用户界面由一个文本字段组成，用户可以在该字段中输入由一个或多个关键词组成的查询。当用户清楚地知道自己在寻找什么，并且能够使用与索引相同的术语来表达他们的查询时，基于关键词查询的搜索引擎就能很好地工作。对于我们包含202,868个带有文本描述的项目的多媒体数据库，我们用一个基于类别的界面来补充这样一个搜索引擎，其类别结构是根据数据库的内容量身定制的。这便于浏览，并为用户提供了查找命名实体的可能性，即使他们忘记了它们的名字。我们证明，这种方法允许那些未能回忆起命名实体名称的用户轻松地检索数据。在我们所有的实验中，在类别上进行1次查询，平均点击2.49次，相比之下，在数据库的传统文本搜索引擎上进行5.68次查询时成功率为68.3%，或者当用户也使用谷歌时进行6.01次查询，成功率为97.1%。
