Development of cognitive model for navigating on the web

Abstract

The objective of this thesis is to build a cognitive model of human performance in Web-assisted tasks. The research is driven by the following questions: What are the most important factors in determining success in Web-assisted tasks? What cognitive mechanisms are involved in these factors? What kind of Web navigation support can be conceived based on the knowledge gained from the previous questions?

The approach is based on the simultaneous consideration of theory, method and real-world applicability. Web navigation is grounded in theories of Cognitive Science (Text Comprehension in particular), and Information Science (Human-Computer Interaction in particular). Experimentation, statistical analysis and modeling are conducted. Practical needs of Web engineering are taken into consideration.

This research investigates how real Web applications are used. A sequence of repeated studies shows that a combination of two factors is the most important determinant of human performance in Web-assisted tasks: a structure-related factor (spatial ability) and a content-related factor (domain expertise). Spatial cognition is involved in representing the structure of the information space, while domain knowledge is necessary for understanding and selecting relevant content.

Factors, such as spatial ability and domain expertise, can only be measured with specialized tests, which cannot be implemented in realistic Web applications. For this reason, Web-logging data is used to calculate metrics of Web navigation behavior. Metrics referring to the structure of user navigation are called syntactic, whereas metrics referring to the visited content are called semantic. It is demonstrated that syntactic (structural) metrics indicate users' navigation styles, for example, if they prefer to revisit pages rather than viewing new pages, or if they return to previously viewed pages using the back button or just by following links. Semantic metrics indicate if users are effective in pursuing their goals independent of their navigation styles. These navigation metrics can be used in building user-models for adaptive Web applications such as recommender systems.

A cognitive model of Web navigation (labeled CoLiDeS+) is proposed. Theoretical and empirical arguments are used to motivate the main assumptions of the model which are: (a) users build and update a mental representation of the information space being navigated; and (b) they assess relevance and make decisions to select particular contents based on both prior knowledge they have about those contents, and knowledge they gain from the local context of those particular contents (i.e., what contents they link to). CoLiDeS+, an augmented version of CoLiDeS (Kitajima, Blackmon, & Polson, 2000), uses Latent Semantic Analysis to model assessments of relevance and user navigation history (sequence of selected links) to model contextual information involved in making navigational decisions. This latter feature is the main distinguishing characteristic of CoLiDeS+. The model has been empirically tested for its accuracy in simulating actual user behavior and its utility in generating Web navigation support. It is shown that CoLiDeS+ performs better in modeling user behavior than its previous version (CoLiDeS) and the navigation support generated from its simulations has a positive impact on user behavior and task outcomes.

This thesis advances the scientific understanding of human performance in knowledge-intensive tasks and contributes to designing useable and accessible information environments.