From Rational Agent to Human with Bounded Rationality

Abstract

To clarify and address the errors that occur in model parameter estimations and behavior predictions, researchers may need to start with investigating the hidden gaps between rational agent and human that are ignored or covered by oversimplified model assumptions. These gaps could occur in both factual, ad hoc retrieval and whole-session interactive retrieval and involve multiple aspects of search interactions, including not only user characteristics and their search strategies but also search task features, search interfaces, as well as situational factors. In this chapter, we summarize and briefly discuss the gaps we identified between simplified rational assumptions and empirically confirmed human biases and then propose a preliminary bias-aware evaluation framework to describe the connections between different stages of search sessions and diverse types of biases. The identified gaps will serve as the basis for developing bias-aware user models, search systems, and evaluation metrics.