The Search Performance Evaluation and Prediction in Exploratory Search

The exploratory search for complex search tasks requires an effective search behavior model to evaluate and predict user search performance. Few studies have investigated the relationship between user search behavior and search performance in exploratory search. This research adopts a mixed approach combining search system development, user search experiment, search query log analysis, and multivariate regression analysis to resolve the knowledge gap. Through this study, it is shown that exploratory search behavior could be featured with long query length, continuous query reformulations, careful search results evaluation and numerous search results click-through. Moreover, a search interaction-performance model has been constructed using multivariate regression analysis. The model evaluation indicates that the search interaction-performance model could effectively evaluate and predict user search performance in exploratory search. Meanwhile, the study has identified two main exploratory search strategies that could guide searchers into the right information: breadth-first and depth-first search strategies. The search interaction-performance model indicates that the breadth-first strategy tends to achieve better search performance than the depth-first strategy. Finally, since search interactions could affect search performance significantly, a real-time interactive search system has been proposed to guide users through different search trails and provide real-time feedback for search interactions in order to achieve better search performance