How Cognitive Computational Models Can Improve Information Search

This chapter discusses why and how a computational cognitive model that captures the broader set of processes of information search is important and useful. The first reason why this can be useful is when information search is only one component of a broader task. A better understanding of the broader process of information search can lead to better metrics of relevance that are specific to the broader task in which the user is engaged. The second reason is that it helps to develop better personalized tools that are more compatible with the individual users as they search information for different purposes. Two examples of such computational cognitive models are presented. The first model, SNIF-ACT, demonstrates the value of adopting a theory-based mechanism, called the Bayesian satisficing mechanism (BSM), that selects information search strategies based on ongoing assessment of the information scent cues encountered by a user as he or she navigates across Web pages. The second model, ESL, tracks both learning of knowledge structures and search behavior in a social tagging system over a period of eight weeks as they continuously search for Web documents. These computational cognitive models generate explicit predictions on what users will do when they interact with different information retrieval systems for different tasks in different contexts. Computational cognitive models therefore complement existing computational techniques that aim to improve information or document retrieval. At the same time, they allow researchers to develop and test unified theories of information search by integrating the vast literature on information search behavior in different contexts.