Our work initially focused on maps. Map data may be large in size, causing long delays on slow networks and it may include a level of detail, which cannot be interpreted on low-resolution displays. Vector map data lends itself to partial presentation and processing of different elements. We were able to use this to allow the prioritisation of data according to its semantic types, eg roads, rivers, buildings, distance from location of interest; and its properties, eg different scale representations of the same feature. In this way mode of transport affects the priority of representations of roads, railways, footpaths etc. Speed affects inclusion of distant or small features. Task (work, leisure etc.) affects inclusion of tourist symbols. We also describe limits over managed resources in order to constrain the selection. Screen resolution and available bandwidth affect the combination of data selected, in order to meet specified goals: time to download may depend on speed, drawing complexity may depend on vibration. Note that the selection of data to display considers the combined effect of all the data in the selection, rather than processing data on an element-by-element basis.
Contextual mediation is illustrated in the Figure. The left-hand map shows the complexity resulting from no mediation. The mediated maps are for navigating in a car at 30 to 40mph, hence the concentration on major roads and the omission of minor roads away from the immediate vicinity (the centre). The data was loaded within a 25s deadline over a simulated GPRS network. The difference between the tourist and workers maps is subtle, but helps the worker to scan the map without distraction. Where features have hyperlinks from them, eg describing tourist attractions or access to delivery points, the ease of selecting links also improves. In tests we found the resulting maps to be more predictable in their download time than unmediated maps. Predictable delays have been shown to be a key factor in user satisfaction in the web. The omission of unnecessary detail was shown to provide improved clarity and faster navigation in user tests.
Our approach is applicable to other semantically rich data intensive applications. We are now investigating applying this approach to collaborative tools such as message services, shared whiteboards, engineering and maintenance plans and web based documents. In collaboration with the University of Southampton we have been investigating the mediation of hyperlinks and the use of mediation in pervasive information systems. There is also related work starting on Ubiquitous Computing for Healthcare in the Community, which will be focusing on monitoring of patients with clinical conditions as they go about their normal activity.