Automatically generating personalized user interfaces with Supple
Download 5.78 Mb. View original pdf
|
| Fig. 8. Three types of transitions between elements of a user interface illustrated with respect to the A / V interior node entering (ent), when a descendant of A / V node is manipulated following an element that is not its descendant sibling switching (sw), when user manipulates two different descendants of the A / V node and leaving (lv), when a user manipulates a descendant of A / V and then navigates to an element outside of the A / V sub-tree. We start by defining $ to be of the form: $ R ( S f ), T ≡ T ∈T | T |− 1 i = 1 N ( R , e i − 1 , e i ) + M R ( e i ) (1) where N is an estimate of the effort of navigating between widgets corresponding to the subsequent interface elements, e k ∈ S f , referenced in a trail, T , and M is a manipulation cost function that measures how good each widget is for manipulating state variables of a given type. Hence, the cost of a rendering is the sum of the costs of each user operation recorded in the trace. Eq. (1) satisfies the first of the three requirements, but requires reanalyzing the entire user trace each time anew cost estimate is necessary, and it fails to satisfy the remaining two requirements. To address those limitations, we first define N : { sw , lv , ent } × W c → to be a function, specific to container widgets, that reflects the cost associated with navigating through a rendered interface. In particular, there are three ways (denoted sw , lv, and ent) in which users can transition through container widgets (Fig. 8). If we consider a container widget w repre- senting an interface element e, the three transitions are entering ( ent ) , when a descendant of e is manipulated following an element that is not its descendant sibling switching ( sw ) , when user manipulates two elements that belong to two different descendants of e; and leaving ( lv ) , when a user manipulates a descendant of e and then navigates to an element outside of the e sub-tree. For different types of container widgets, these three transitions are predicted to increase user effort indifferent ways. For example, suppose that e is rendered with a tab pane widget. Then N ( sw , e ) , which denotes the cost of switching between its children, would be high, because this maneuver always requires clicking on a tab pane. Leaving a tab widget requires no extra interactions with the tab. Entering a tab pane usually requires extra effort, unless the tab that the user is about to access has been previously selected. In the case of a popup window, both entering and leaving require extra effort (click required to popup the window on entry, another click required to dismiss it) but no extra effort is required for switching between children if they are rendered side-by-side. Recall that our interface specification is a hierarchy of interface elements. Assuming a rendition where no shortcuts are inserted between sibling branches in the tree describing the interface, one can unambiguously determine the path between any two elements in the interface. We denote the path between elements e i and e j to be p ( Download 5.78 Mb. Share with your friends:
|