| When an interactive search system returns a results list that fails to meet its user's information need, the user experiences a query failure. With the present generation of search systems, except for the most common and simple information needs, users often encounter query failure. This dissertation examined the behaviors searchers use when overcoming query failure. Specifically, this dissertation compared searches conducted on systems operating at three different levels of performance in a single mixed-model factorial experiment, with system performance as the independent variable. The General Linear Model and planned contrasts were used in an exploratory analysis of the effects of system performance on system responses, search behavior, and searcher productivity. Thirty-six volunteers from the Rutgers University community participated in the study. The study found that when system performance is degraded, searchers increase the pace of query submissions. Inter-query time intervals are shorter when results lists are shorter and when a spelling error message is displayed with a results list. These findings suggest that a system capable of monitoring a user's query submission rate and the characteristics of its own responses may be able to detect and assist a user experiencing a difficult search. |
