Simulated cognitive topologies: automatically generating highly contextual maps for complex journeys

Abstract

As people traverse complex journeys, they engage in a number of information interactions across spatial scales and levels of abstraction. Journey complexity is characterised by factors including the number of actions required, and by variation in the contextual basis of reasoning such as a transition between different modes of transport. The high-level task of an A to B journey decomposes into a sequence of lower-level navigational sub-tasks, with the representation of geographic entities that support navigation during, between and across sub-tasks, varying relative to the nature of the task and the character of the geography. For example, transitioning from or to a particular mode of transport has a direct bearing on the natural level of representational abstraction that supports the task, as well as on the overall extent of the task’s region of influence on the traveller’s focus. Modern mobile technologies send data to a device that can in theory be context-specific in terms of explicitly reflecting a traveller’s heterogeneous information requirements, however the extent to which context is explicitly reflected in the selection and display of navigational information remains limited in practice, with a rigid, predetermined scale-based hierarchy of cartographic views remaining the underlying representational paradigm. The core subject of the research is the context-dependent selection and display of navigational information, and while there are many and varied considerations in developing techniques to address selection and display, the central challenge can simply be articulated as how to determine the probability, given the traveller’s current context, that a feature should be in the current map view. Clearly this central challenge extends to all features in the spatial extent, and so from a practical perspective, research questions centre around the initial selection of a subset of features, and around determining an overall probability distribution over the subset given the significance of features within the hierarchically ordered sequence of tasks. In this thesis research is presented around the use of graph structures as a practical basis for modeling urban geography to support heterogenous selections across viewing scales, and ultimately for displaying highly context-specific cartographic views. Through an iterative, empirical research methodology, a formalised approach based on routing networks is presented, which serves as the basis for modeling, selection and display. Findings are presented from a series of 7 situated navigation studies that included research with an existing navigation application as well as experimental research stimuli. Hypotheses were validated and refined over the course of the studies, with a focus on journey-specific regions that form around the navigable network. Empirical data includes sketch maps, textual descriptions, video and device interactions over the course of complex navigation exercises. Study findings support the proposed graph architecture, including subgraph classes that approximate cognitive structures central to natural comprehension and reasoning. Empirical findings lead to the central argument of a model based on causal mechanisms, in which relations are formalised between task, selection and abstraction. A causal framework for automatically determining map content for a given journey context is presented, with the approach involving a conceptual shift from treating geographic features as spatially indexed records, to treating them as variables with a finite number of possible states. Causal nets serve as the practical basis of reasoning, with geographic features being represented by variables in these causal structures. The central challenge of finding the probability that a variable in a causal net is in a particular state is addressed through a causal model in which journey context serves as the evidence that propagates over the net. In this way, complex heterogeneous selections for interactive multi-scale information spaces are expressed as probability distributions determined through message propagation. The thesis concludes with a discussion around the implications of the approach for the presentation of navigational information, and it is shown how the framework can support context-specific selection and disambiguation of map content, demonstrated through the central use case of navigating complex urban journeys.