Supporting document use through interactive visualisation of paragraph-level metadata.

This work examines the thesis that the interactive visualisation of paragraph-level metadata can be used to increase the efficiency of goal-directed search in corporate documents. The design of a tool to support goal-directed search via the visualisation of paragraph-level metadata is explored, defined and implemented. A new evaluation method, named best-case analysis, is developed and applied to determine the potential utility of this tool. A new visualisation, named GridVis, is described which displays a document's paragraph-level metadata. The design decisions are supported by task analysis, evaluation and a review of the information visualisation literature. A new technique based on interviews with users is developed to create metadata and metadata taxonomies that will service common information needs. An interactive prototype of GridVis is subjected to an experimental performance evaluation. A qualitative analysis shows that performance could be improved if users made more use of the information provided by GridVis. The only quantitatively significant result was GridVis' lower task time which prompted suspicion of GridVis's utility. GridVis was re-implemented to realise the design ideas inspired by the qualitative analysis and user feedback. A new evaluation method, named best-case analysis, is described. It makes use of GOMS and information foraging theory to build a model that computes a 'best-case' performance. This is then compared to the benchmark performance of twenty four domain experts using paper versions of the same documents. This evaluation method reveals that GridVis potentially offers either higher recall or faster task times than paper, but not both. It shows that the time saved by avoiding irrelevant paragraphs is largely consumed by the extra navigation time required. If however the design were improved to make it 50% faster to use, it would give performance that is broadly superior to that of paper