Navigation in degree-of-interest trees
Degree Of Interest (DOI) trees are a focus+context method of information visualization, in which a tree-like information structure is displayed with different levels of detail depending on the user's degree of interest in each item (Card & Nation, 002; Furnas, 1981). We describe an ACT-R model for how people search for a target item in a DOI directory-like structure. Given a search goal, the user must click on those nodes in the tree under which the search goal may have been classified. The model has two components: a visual search component that selects which nodes (from those displayed on the screen) must be attended and in which order, and a semantic component, that evaluates all the attended nodes and selects the node to be clicked on next. The visual search is roughly based on Logan(1996)'s CODE theory of visual attention and essentially groups nodes that are close together (in physical distance) into ``blobs''. A blob is first selected to be attended, and then all the items in the blob are sequentially attended and evaluated. The semantic component of the model evaluates the nodes based on its semantic similarity with the target and prefers nodes that are hypernyms of the target item. We use GLSA PMI-based (Matveeva, Farahat & Royer, 2005) computations to estimate semantic similarity; for each search item, we also use category classification taken from WordNet (Fellbaum, 1998). We explore other solutions for similarity computations (LSA --- Landauer & Dumais, 1997, LSA/GLSA combined with elaborations). In this talk I will present the current variant of the model and discuss some of the challenges that we encountered, related both to visual search assumptions and also to semantic similarity.
Budiu, R. ; Pirolli, P. L. Navigation in degree-of-interest trees. Proceedings of the Twelfth Annual ACT-R Workshop; 2005 July 15-17; Trieste; Italy.