Automatic acquired domain knowledge for ad hoc search: evaluated results

The automatic acquisition of usable domain knowledge is a challenging issue. Such knowledge can be employed to assist a user in searching a document collection. This can be done by suggesting query modification options based on the knowledge uncovered by analyzing the document collection. We acquire such knowledge by simply exploiting the documents´ markup structure. This gives us a domain model tailored to the particular collection. But how good is such a model? We present results of two evaluations. The first one looks at the actual domain model. We will discuss how users judged the relations encoded in the model. The second evaluation is task-based and goes a step further. It investigates how a search system that applies the automatically constructed domain model performs compared to a standard search system.