To convey the structure of this thesis, each chapter is described very briefly at first and
summarised by the thesis map in Figure 1, the contents of each chapter are then
described in more detail. Chapter 2 reviews the literature in the fields related to this
work. It describes relevant background material and lays out the evidence to warrant the
choice of techniques used later in the thesis for design and evaluation work. Chapter 3
reports two exploratory studies of document use in a corporate environment which
sought to assess the feasibility of using paragraph level metadata. Chapter 4 describes the design of GridVis, an interactive visualisation of paragraph-level metadata to
support goal-directed search. Chapter 5 reports user trials, task analysis, and
experimental evaluation that were used to assess and further refine the design of
GridVis. Chapter 6 describes the best-case analysis used to ascertain the potential performance of the interactive visualisations of paragraph-level metadata as embodied
by GridVis. Chapter 7 reports the results of this evaluation, and Chapter 8 ties together the conclusions and contributions of this research and suggests avenues for further
work. Figure 1 illustrates the way each of these chapters fit into the thesis as a whole, and locations of the various contributions enumerated in section 1. The content of each
Chapter 1 section 2 33
G
Ch 1 - Introd uctionBackground investigation Evaluation
Ch 2 - B a ckground Literature
M eta data literature Inform ation seeking Inform ation visualisation Inform ation foraging Evaluation m ethods Related w ork
Ch 3 - C o rp orate d oc um ents and how m etadata could fit in
W riter study
interview s 7 p a rticip a n ts
R eader study
inte rview s 4 pa rticip a n ts C h a ra c te risa tio n of d o c u m e n ts u s e d C h a ra c te risa tio n of d o c u m e n ts su itab le for e n ric h m e n t with p a ra g ra p h -le v e l m e ta d a ta Investigation Into n u m b e r th e g o a ls th a t a re m e ta d a ta e x p re ssib le Design a nd prtxluction
Ch 4 - G ridV is C o nstru ction
Selecting the docu m en ts M eta data Design and Production M eta data Taxo nom y D esign and P roduction
Interview with 8 p a rticip a n ts, k P ro v id e d inform ation o n inform ation
n e e d s a n d w hat is n e e d e d to d isc n m in a te rele v an t m a teria l
Selecting the docu m en ts V isualization Design
Ch 8 - C o n clu sio n s and further w ork
C o nclusions and contributio ns Further w ork_____________________
Ch 5 - Initial e v a lu atio n
U ser trial
4 p a rtic ip a n ts • 3 re a d in g ta s k s with GridVis (overview , g o a l-d ire c te d s e a r c h a n d sorting)
T a sk analysis E xp erim e nta l e va luation
12 p a rticip a n ts - G ridVis v s. S c rolling w indow - within s u ^ e c t s d e s ig n
E xpert ju d g e ratin g s for p a ra g ra p h r e le v a n c e
Ch 6 - A best-case ana lysis
^ ^ e s t - c a s e ana lyses in the abstract
T h e p r o 's a n d c o n 's o l b e s t- c a s e a n a ly s e s
he best case ana lysis of G ridV is ^ S t r a t e g y - le v e l defin ition
S tudy to identify ta g s c e n t s tra te g y (2 p a rtic ip a n ts se le c tin g ta g s with in s tru m e n te d v e rsio n o l GridVis)
P e rform an ce-level definition
G O M S m o d e ls lo r G ridVis u s e
% M odels o l e x p e r ie n c e ■ U n e fitting to d a ta from d o c u m e n ts a n d P a p e r P e r fo r m a n c e E x p erim e n t
T he ben ch m a rk - P aper p e rform ance expe rim en t
24 su b je c ts . G o a W ir e c te d s e a r c h with p a p e r d o c u m e n ts . R etrieval p e rf o rm a n c e a n d p a ra g ra p h re a d in g tim e s m e a s u r e d u s in g v id e o f o o tag e .
M a te r ia ls
D o cu m en t s e le c tio n a n d m anipulation E x p ert ju d g e r a tin g s for p a ra g ra p h r e le v a n c e Q u e ry c o n sfru ctio n
P erform ance m etrics
Ch 7 - R esults fro m th e best-case a n a lysis
/ ^ e b e n chm a rk p e rform ance
N ^ ^ n a l y s i s of high le vel re a d in g str a te g ie s V A nalysis of ta s k p e rfo rm a n c e
G ridV is Vs B e nchm ark perform ance
Figure 1 Thesis Map. The location o f each major
,-,Q)
and m inor%
contributionC hapter 2 review s literature from the various areas related to this research. Section 2 on m etadata describes docum ent m etadata from different traditions, looks at som e o f the issues surrounding its production, and review s the w ays it has been used to support search w ithin docum ents. A section on inform ation visualisation, section 3, review s its various techniques and fields evidence to w arrant the design choices in the developm ent o f G ridV is. The follow ing section, section 4, draw s on theories o f inform ation seeking to argue for the potential utility of m etadata taxonom ies, and uses evidence for the im portance o f goal-directed search in the w orkplace to argue that it is w orth supporting and notes the continuous prevalence o f paper docum ent use. The next section, 5, looks at evaluation m ethods. It exam ines the em pirical m ethods and m etrics used to evaluate the support o f goal-directed search w ithin docum ents. A range o f theoretical evaluation
methods are described and the particular choice of techniques used in this research is
justified. One of the more recently developed of these techniques, the use of information
foraging theory, is the subject of the following section, 6, where its foundations, use and
validity are discussed. The final section, section 8, covers visualisation work related to
the work in this thesis, it is critically assessed and the inadequacies of evaluation
techniques which might substitute for the best-case method developed in this research,
are exposed.
Chapter 3 reports two exploratory studies of document use in the head office of a large
UK retailer which attempted to examine the feasibility of using paragraph-level
metadata to support goal-directed search within corporate documents. A subset of the
documents used are identified as being suitable for enrichment with metadata. Their characteristics are described. An analysis of the goals workers bring to documents they
read examines the degree to which these goals can be represented with metadata made
up of keywords or noun phrases.
Chapter 4 describes the design and construction of GridVis, an interactive visualisation
of paragraph-level metadata. The techniques used to construct the high quality metadata needed are described in section 2. These include techniques adapted from previous
research and a novel technique developed in this research, to elicit the information used
to discriminate relevant material. The design of GridVis is described in section 3; the
contribution to the design, of evidence from numerous rounds of evaluation, is
discussed.
Chapter 5 reports a user trial, task analysis and experimental evaluation of GridVis. In
section 1, the description of a user trial of an early static mock-up of GridVis supports
discussion of a number of design problems and highlights methodological concerns. A task analysis of the use of paragraph-level metadata for goal-directed search is then
Chapter 1 section 2 35
reported for use in a review of the GridVis design (see section 2). The final section of
this chapter, section 3, describes the evaluation of a full implementation of an improved
design, using a controlled experimental comparison with a simple scrolling window.
Results from this experiment are used to argue for the need for an evaluation
methodology which can assess the ultimate potential of a class of designs.
Chapter 6 describes the best-case analysis method and its application to the interactive
visualisation of paragraph-level metadata. The best-case analysis method is described in
section 1.1. Its general advantages and limitation are discussed. In section 2 the best-
case model of GridVis use which incorporates information foraging theory is described.
First, in section 2.1, the model is described at a high, or strategy level. Then, in section
2.2, it is described at a computational, or performance level. Section 3 describes an experiment to measure human goal-directed search performance with paper documents.
Chapter 7 reports the results of the best-case analysis. The results from the experiment on performance with paper are reported first in section 1. The navigational strategies
used with paper are described and used to argue the validity of comparing the
performance on paper with that computed by the best-case model of GridVis. The best-
case performance of GridVis is compared with that of paper in section 2. The relative advantages of various affordances offered by GridVis are reported and an assessment is
made of the general viability of interactive visualisations of paragraph-level metadata.
Chapter 8 summaries the contributions and conclusions offered by this research and describes various avenues for further work.