Interactive TREC-4 at Georgia Tech

At Georgia Tech, we investigated the eeectiveness of a visualization scheme for Information Retrieval systems. Displayed like a bar-graph, the visualization tool shows the distribution of query words in the set of documents retrieved in response to a query. We found that end-users use the visualization for two purposes: to gain speciic information about individual documents { such as the distribution of diierent query words in that document. to gain aggregate information about the query result in general { such as getting a sense of the direction of the query results. In general they used the visualization tool as much as the title and full text in the process of deciding if a document addresses the given search topic. In structured post-session interviews with searchers, we also obtained information about what the searcher liked, what was frustrating to them, and what they wanted in the system.