FLEXIBLE HUMAN SERVICE INTERFACES
Josef Spillner, Iris Braun and Alexander Schill
Chair for Computer Networks, TU Dresden, 01062 Dresden, Germany
Keywords:
WSGUI, Web Services, GUI generation, MDA techniques.
Abstract:
Dynamic web service invocation without special client software may help the adoption of service-oriented
architectures on the consumer stage. Ad-hoc usage of services requires a powerful set of concepts to visu-
alise the service input and output messages in a user-friendly, ergonomic and extensible way. Such concepts
are collected in a research effort named Web Service Graphical User Interface and are presented in the pa-
per in combination with an algorithm to combine the concepts into one imaginary GUI creation engine, for
which a proof-of-concept implementation exists. Extensibility is achieved by using implicit and explicit GUI
generation hints in addition to inference mechanisms based on the message structures.
1 MOTIVATION
WSGUI, or Web Services Graphical User Interface,
describes concepts on how to enrich data model
schema information in the context of web services so
that it becomes usable for humans in interactive sce-
narios. The core idea of WSGUI is that dialogs for
input and output of structured parameters can be gen-
erated dynamically. To achieve usable and appealing
GUIs, the formal message specification will almost
never suffice, hence there is a need for some addi-
tional data on how the GUI should be created. A col-
lection of concepts is presented which is believed to
achieve good overall quality without too much effort
in the creation of the additional data. There might be
other combinations of concepts, unknown as of yet,
but the ones presented here seem to be a good choice
given their properties: First, they follow the paradigm
of model-driven GUI design, making them suitable
for changes in the message parameters where the mes-
sage schema will take precedence over the GUI anno-
tations. Second, they can be used as needed, with few
interdependencies and the ability to incrementally im-
prove the generated GUIs without having to interfere
with the service functionality.
The term GUI shall be used in a wider sense and
will in most cases refer to generated dialogs or masks
for input and output of data.
The nature of web services in practice is that
they’re often described only syntactically, using
WSDL files which include schema information to de-
scribe the messages. This requires a mechanism to
make up for areas where the description lacks by
means of additional GUI generation hints.
WSGUI can therefore be thought of as a family of
different concepts which, when combined properly,
can lead to decent GUIs which are generated auto-
matically while still preserving an appealing look and
feel. Platform independence, integration with the sur-
rounding environment and consistency are further ad-
vantages of employing WSGUI concepts in applica-
tions.
The order of the sections of this paper is such that
the concepts are presented first, followed by some
information on how they might interdepend on each
other. The introduction is complemented by a report
on progress since the initial presentation of WSGUI.
An implementation project and some related work is
presented afterwards.
79
Spillner J., Braun I. and Schill A. (2007).
FLEXIBLE HUMAN SERVICE INTERFACES.
In Proceedings of the Ninth International Conference on Enterprise Information Systems - HCI, pages 79-85
DOI: 10.5220/0002366000790085
Copyright
c
SciTePress
2 WSGUI CONCEPTS
Concepts are introduced here as abstract steps which
are part of the transformation process from the web
service message structure to the GUI. All concepts are
realised through specific algorithms and filters as de-
scribed later. The software performing these steps is
named WSGUI engine, independent of how or where
it is actually implemented. For the GUI, any declar-
ative format can be the target of the transformation
process. WSGUI does not define a new abstract UI
description.
A number of concepts were present in the origi-
nal WSGUI proposal (Kassoff et al., 2003). Some of
them were found not to be needed anymore, while at
least two were found to be crucial so they’re included
in the following list. The others are the result of recent
work performed in this area by the authors. Reasons
for why to discontinue some of the concepts will be
given later.
The two mentioned original concepts are web
service description and form components, and can
be used by software implementations by reading in
GUIDD files, that is, GUI Deployment Descriptors
(Kassoff et al., 2006). New concepts added by the au-
thors include a strong inference mechanism, MIME
awareness, the user-toggle concept and templates.
Each of these concepts helps in shaping a better
GUI while still preserving the aforementioned prop-
erties. All of them are going to be presented in this
section.
2.1 Web Service Description
While most of the WSGUI concepts deal with individ-
ual service messages, being either input or output of
an operation, some of them deal with the global scope.
Navigating to the message requires presentation to the
user, too. This includes the selection of the service
and operation and passing this information to the user.
Such user interface elements include the internation-
alised representations of name, short description and
comprehensive help of both the services and their op-
erations, and also several details such as the caption
of the submission button for input forms.
While WSDL files can contain information such
as help texts, service providers usually do not trans-
late these texts. Especially for WSDL files generated
automatically by reflection from object code or pars-
ing of source code, it becomes necessary to supply ex-
ternal texts to overcome this limitation. The GUIDD
file format provides the
operations
section for the
task of including an arbitrary number of translated
human-readable texts.
2.2 Inference
Dynamic invocation of web services without any ad-
ditional GUI generation hints can still result in differ-
ent user experience, depending on how close the GUI
represents the message structures.
Instead of just providing simple form controls, say
string input fields, for all schema elements, a dynamic
invocation engine can already infer that some data
types require special treatment. For example, if a type
is restricted by an enumeration of values, only those
values should be displayed in a combo box, instead of
letting the user input arbitrary values. Likewise, an el-
ement which may occur n times as a child of its parent
element can be displayed as a list which may be en-
larged or shrunken within the occurrence constraints.
In addition, the XML Schema embedded into
WSDL files might provide valuable initial help texts
for each form control through its annotations. For the
captions of the controls, only the parameter name can
be used, which will not be suitable for human dis-
play unless they adhere to certain conventions (Steele
et al., 2005).
A still largely unexplored area is the coupling of
data types to semantics to generate widgets such as
password edits or multiline edits. None of them can
be inferred automatically. They are all based on the
string datatype as is the generic input widget, there-
fore form components are required in the meantime
to force their usage. Semantic annotations provide
implicit GUI hints, whereas form components are an
explicit concept. How to make use of semantic on-
tologies and how to address usability concerns is out
of scope for this paper and dealt with in other publi-
cations (Khushraj and Lassila, 2005).
2.3 Form Components
The main part of the GUIDD file format is about spec-
ifying abstract form components, each of which is
wrapping a specific GUI control, which can be used to
override decisions otherwise being made by the GUI
control inference mechanism. The GUI control fam-
ily currently proposed for GUIDD is XForms (Boyer
et al., 2006), which does however not imply any usage
thereof in the final generated GUI.
Based on the message schema, both model and in-
stance addressing can be used to locate the position-
ing of the controls, with the latter one overriding the
former since complex data types can be in use mul-
tiple times. The need for instance addressing, which
was not present in the original GUIDD specification,
stems from the fact that due to type definitions be-
ing either named or anonymous, there can be syntac-
ICEIS 2007 - International Conference on Enterprise Information Systems
80
tical ambiguities in the model which however lead to
no difference in the instance. For example, schema
elements might appear as anonymous child elements
in the schema XML, or they might appear as named
reusable types, but the instance XML will be the same
in both cases. Both addressing schemes use higher-
level XPath locators.
In case the instance data is restricted by an enu-
meration and this enumeration is likely to change or
even created dynamically, the concept of XForms fil-
ters has been introduced. This replaces the previously
suggested method of dynamic enumerations, and is
also useful for handling translations. The GUIDD
specification (Kassoff et al., 2006) contains advice on
how to handle filters. In the case of enumerations, it
ensures that instance data not yet covered by a select
item in the form component’s control gets added to
the control automatically, while in the case of trans-
lations, all translations which are not requested from
the engine will be dropped before delivery of the doc-
ument to the client.
2.4 Mime Awareness
Especially for output formatting, when receiving raw
data in base64 or hexadecimal encoding, a MIME
type can be specified to automate the display or other
further use of the data. Since this is only part of
WSDL on a message part level, but not available per
schema element, WSGUI needs to provide a way to
enhance the schema with this information. Some
MIME types might resolve to text effectively, for
example
application/pgp-signature
, but others
might lead to output format specific areas, such as
img
for
image/*
in a web context.
Recently, the proposed XForms 1.1 specification
gained support for describing rich output formats for
its
output
control. Eventually, MIME awareness
might thus be realised using GUIDD’s form com-
ponents sections. However, to maintain backwards
compatibility, there is the OutputTypes section where
MIME types can explicitly be assigned to output form
controls.
2.5 User-toggle Concept
Since for some non-determinable widgets such as the
ones outlined above (password edit and multiline edit)
no automatic choice can be done, the inference mech-
anism would always return a generic input control.
In such a case, unless a form component is available,
the user should have the last word in these cases and
be able to override the inference mechanism’s choice.
This can be achieved by displaying a widget type-
toggling button beside the widget so that it can change
its type to whatever alternative might be available, and
does so by the user’s choice.
2.6 Templates
Form generation results in free-standing forms, con-
trary to the practical need of embedding such forms
into host documents such as web pages with custom
surrounding graphics. Templates make it possible to
define template pages with defined positions for the
generated forms, and let the WSGUI engine output
the filled-in template instead of only a free-standing
form.
This concept is a web-specific extension to the
GUIDD format, as its usage is limited to web environ-
ments. However, experimental work has led to good
form integration with WSGUI templates and there-
fore environment-specific GUIDD profiles will likely
result from this work.
3 CONCEPT
INTERDEPENDENCIES
Now that the existing WSGUI concepts are deter-
mined, an algorithm is presented which is considered
a good compromise between automatic GUI genera-
tion and user-designed UIs. Depending on the efforts
for the creation of the GUIDD file, the result can be
either a generic and visually clumsy user interface or
a highly unique service-specific UI, and in most cases
will be in between.
3.1 Form Controls Generation
The proposed algorithm consists of a single recursive
function,
render()
, which takes as its parameter a
schema element, starting with the top-level element
of the complex schema tree. For each schema param-
eter, the best choice is selected for its representation
in the user interface, and the chosen UI element is
stored into a list as associated with the parameter. For
simplicity reasons, schema elements and attributes are
both treated as elements in the algorithm description.
FUNCTION rend er ( element ) :
IF is complex ( element )
FOREACH chil d element OF element
ren der ( child element )
ELSE
IF formcomponentfound ( element )
f i l t e r enumeration ( )
f i l t e r t r a n s l a t i o n ( )
ui = applyformcomponent ( )
FLEXIBLE HUMAN SERVICE INTERFACES
81
ELSE
IF mimeinfo found ( element )
ui = mimewidget ( )
ELSE
f i l t e r documentation ( )
IF inf ere nce po s si b le ( element )
ui = in fer re d co n tr o l ( )
ELSE
ui = generic c on t ro l ( )
user t og gl e ( )
s tor e ( child element , ui )
The algorithm describes how to iterate over the
whole schema, recursively in the case of complex el-
ements which have child elements, to find out all ele-
ments. For each of them, it is decided whether to use
explicit GUI information in the form of form compo-
nents if they are available, or to apply implicit hints
such as MIME awareness or possibly other semantic
annotations. If a form component can be located, the
contained form control must pass some filters, as is
implied by the form components concept. If neither
explicit nor implicit information is available, infer-
ence is the fall-back mechanism, which in some cases
needs to add a user-toggle button to the side of a con-
trol.
Most concepts are present in this algorithm. The
remaining ones deal with web service-specific infor-
mation such as operations and are handled outside the
scope of the creation of the form controls themselves.
The algorithm should instead be part of a pipeline
which lays out the list of form controls according to
how the GUI should eventually look like.
3.2 Pipelining the Concepts
While the presented algorithm transforms message
parameters into form controls, it barely covers lay-
out and geometry settings, that is, how to place the
controls onto the resulting GUI. The only way of in-
ferring those settings is to either build up a tree-like
view, with complex data types being containers for
simple types being represented as controls, or to use
generic schema-based GUI element positioning algo-
rithms as explored in (Kruschinski, 1999).
Since there is no predominant way of specify-
ing layouts, WSGUI does currently not include any-
thing else but an implementation-dependent tree lay-
out convention. The tree-like structure of the mes-
sages the forms are based on suggests that this is an
acceptable resolution.
The usage of explicit layout and geometry infor-
mation could well be seen as a second transformation
step in a pipeline, with the algorithm presented above
being the first step. A third (and last) step would be to
integrate resulting forms into host documents, as de-
Figure 1: WSGUI engine as a pipeline.
scribed in the WSGUI templates concept entry. Fig-
ure 1 describes how the pipeline works. It is note-
worthy that not only the schema, but also related con-
text information such as operation-specific informa-
tion (name and submission URL) and the position in
a complex process, if any, are used to determine the
generated GUI.
4 COMPARISON TO THE
ORIGINAL WSGUI CONCEPTS
In 2003, the first paper on WSGUI was published
(Kassoff et al., 2003), which outlined the basic ideas
present in current-day WSGUI research. However,
with other standards progressing fast, in particular
those in the area of XML, it was time to reconsider
the ideas of the paper and re-evaluate them for prac-
tical usage. The ideas will be listed, described and
evaluated in the forthcoming paragraphs.
Form Components as specified in a GUIDD file
are still considered a necessary WSGUI concept. Sev-
eral subtle differences to the original idea exist, such
as using XForms controls directly (apart from fil-
tering) as opposed to merely borrowing from the
XForms syntax. The central idea of form components
is explicit assignment of visual controls to parts of the
data model. The reliance on XForms is discussed in
the section about the form components concept and
helps keeping WSGUI lean.
Form Navigation is an interesting concept which
can morph the UI around a selection between di-
rect editing, item selection and constraint-based query
views. The intended implementation looks doable,
ICEIS 2007 - International Conference on Enterprise Information Systems
82
nevertheless this feature is not included in the cur-
rent research since it requires additional work from
the authors of GUIDD files. The new concept of user-
toggle is a similar one, which stresses the choice on
the part of the user.
Variable-sized Lists is definitely needed for web
services which contain schema elements of varying
quantities. However, it was found that WSGUI en-
gines can automatically generate GUIs which take
care of adding and removing controls according to the
allowed range of quantity. For example, XForms pro-
vides the xf:repeat module which eliminates the need
to copy the functionality in the GUIDD file.
Virtual Operations are best implemented within
processes, and delivered transparently to the client as
part of a WSDL file by a process engine. Standards
such as BPEL, the Business Process Execution Lan-
guage, make it possible to build such virtual opera-
tions without the direct need for WSGUI concepts,
since the elements of a GUIDD file can refer to such
operations as well.
Dynamic Enumeration refers to special web ser-
vice method which return a list of all possible values
of a variable, for display in a non-editable drop-down
box or a similar widget. Value constraints are better
specified explicitly in a schema file as part of a sim-
ple type enumeration restriction. As before, GUIDD
form components can map to these lists of values,
but no special requirement is needed for the service.
XForms filters ensure that if GUIDD and the enumer-
ation instance data go out of sync, the resulting GUI
does still display formally correct choices to the user.
Therefore, the current GUIDD files do not need the
corresponding XML descriptions anymore, and web
services do not need to be enhanced with helper meth-
ods anymore.
Look-and-feel Stylesheets The task of the L&F
stylesheets is to convert an abstract user interface to
a concrete one. The wide adoption of abstract GUI
description languages such as XForms and our im-
plementation work has led us to believe that no such
stylesheets are needed per se as part of GUIDD.
To conclude this section, a number of concepts
could be saved, partially thanks to alternative ap-
proaches leading to similar or better user experience,
and at the same time leading to less work for both the
GUIDD authors and the service providers.
5 RESEARCH AND
IMPLEMENTATION PROJECTS
The failure of adoption of previous efforts similar to
WSGUI led to the conclusion that file formats like
GUIDD would only be deployed if sufficient support
for it is available from software developers. The au-
thors of the paper will present their implementation in
this section, but also refer to others which share some
of the ideas.
5.1 Project Dynvocation
Most of the WSGUI concepts described above are
being implemented as part of Project Dynvocation
(Spillner et al., 2006). The Dynvoker servlet gener-
ates input forms and result pages on the fly from web
services, using inference as heavily as possible and
applying GUIDD hints wherever they are available.
Most of the implementation work was aimed at
creating a servlet which produces XHTML pages with
embedded XForms. Thanks to the use of XForms
in GUIDD form components, the XForms controls
found in the GUIDD form components can be used
natively after the XForms filters have been applied.
As opposed to its predecessor, HTML forms, XForms
offers schema binding and thus also type safety and
separation of data from its presentation.
Its focus on inference makes it possible to invoke
web services without any additional information, and
already provides type safety and flexibility in terms
of changing list sizes and user-toggle actions. Adding
a GUIDD will however greatly enhance the user ex-
perience. An editor for the creation of GUIDD files
has been developed as part of a thesis work of one of
the authors recently. It reuses some of the WSGUI
concepts to make the authoring process as simple as
possible. The decoupling of a web service descrip-
tion file (the contract) and GUI hints (the presentation
layer) greatly helps in authoring and customising the
latter.
However, Dynvoker is not limited to producing
XHTML pages for the interaction with simple web
services. A few extensions will be described now.
5.1.1 Existing Extensions to Dynvoker
The Dynvoker software has been extended to han-
dle complex service interaction scenarios like BPEL-
composed services. The work is called GUI4CWS
(GUI for Complex Web Services) (Bleyh, 2006) and
uses a subset of BPEL in combination with WSGUI
concepts to achieve goal-oriented workflows.
FLEXIBLE HUMAN SERVICE INTERFACES
83
Beside XHTML output, a way to create XSWT
dialogs for the Eclipse platform has also been worked
on. While the XHTML generation works mostly pro-
grammatically, the Eclipse dialog generation is per-
formed by a XSL transformation. However, not all
corner cases could be handled to full satisfaction.
Other projects (Moebius, 2006) also try to generate
GUIs from XML Schema or WSDL files purely by
applying a XML transformation and similarly leave
out the more complicated transformations. It remains
to be seen to which extent the logic behind the GUI
generation can be represented in a declarative trans-
formation document if the goal was to implement all
WSGUI concepts.
5.1.2 Details of the Gui Generation
Before a form is rendered by Dynvoker, the user is
guided to select a web service and an operation. All
of the navigational user interface elements are derived
from a navigation schema as well, thus re-using Dyn-
voker’s ability to generate dialogs for several target
platforms.
Hence, Dynvoker presents itself to the user with
an interface which is automatically generated from
an XML Schema file and the corresponding GUIDD
file. It allows the user to select a service by enter-
ing the URL of a WSDL file, possibly in combination
with a GUIDD file in addition, and some option but-
tons which will influence the generation of the GUI,
mostly to be able to see the difference between using
and not using any of the mentioned WSGUI concepts.
If no GUIDD file is specified, the whole GUI genera-
tion process relies on inference mechanisms based on
the information in the WSDL file.
After submitting the service selection form, the
GUI for the input message of the WSDL file is ren-
dered, and the user again fills out all of the fields.
Another submission action now invokes the service
using Dynvoker as a proxy, and returns the result to
Dynvoker. The results are then rendered as an output
form which is sent to the user.
Architecturally (figure 2), a DynvokerEntrance
object receives the parameters of the call, provid-
ing an abstraction layer between the Dynvoker im-
plementation as a servlet, portlet or library on one
side and the remaining parts on the other side. Dyn-
vokerCore as a state machine then takes over the pa-
rameters and dispatches the call as needed. In most
cases, an XFormsAdapter object will be called which
implements the first step of the WSGUI pipeline.
This includes assigning form components to the in-
dividual parameters of the currently selected web ser-
vice message, calling XFormsFilter whenever trans-
lations need to be eliminated from the form controls
Figure 2: Dynvoker architecture.
or enumerations are transformed. To render a suitable
XHTML page which is to include the generated form
or to include the form into an existing template page,
XFormsHost can be called. This object also handles
the placement of the form controls into a tree layout.
5.1.3 Xforms Filters
As mentioned before, form controls are not eas-
ily suitable for dynamically generated list datatypes,
for which only the control’s name and help could
be given deterministically, while the values might
change. When they don’t change too often or some of
them are present in most invocations, XForms filter-
ing comes to the rescue. The graphics 3 shows how
a partially applicable form control definition is lay-
ered on top of the corresponding auto-generated form
control.
XPath expressions are used to match the related
XML nodes of both controls. Similar to how form
controls themselves are referenced using XPath ex-
pressions, name spaces are preserved and handled
correctly, making it a suitable mechanism even for
complex schemas.
The GUIDD extension proposed in the GUI4CWS
document for handling list types by introducing a
placeholder
<items/>
tag is not necessary anymore
thanks to the filtering concept.
Filters are also used in Dynvoker to hide all trans-
lations which are not requested by the client in la-
bels, help texts and hint texts. Additionally, they’re
also used to use existing XML Schema annotations as
help texts if no help text has been given for a control
through the GUIDD file.
5.2 Other Wsgui-related Projects
A project similar in its goals is the XML Forms Gen-
erator (Kelly et al., 2006), which is part of the Eclipse
framework. It already supports external translation
ICEIS 2007 - International Conference on Enterprise Information Systems
84
Figure 3: XForms filters.
texts, which matches the GUIDD facility of form
components, but also deals with rendering options. Its
focus on web services would make it an ideal candi-
date for using further WSGUI concepts so that those
issues found in real-world web services and GUI gen-
eration of their messages could be overcome.
An architecture for discovery and dynamic invo-
cation of services has been presented in (Steele et al.,
2005). It focuses on multimodal access more than on
in-depth discussion of GUI generation for visual ac-
cess, which is still the most important kind of interac-
tion between services and humans. Building a service
discovery model on voice recognition is another idea
present in this work, which will further help in ad-hoc
service usage. The presented way of generating the
GUI could profit from employing WSGUI concepts,
as the ambiguity of WSDL parameter naming is ad-
mitted openly in this work. The proposed solution,
to adhere to naming conventions for WSDL message
variables, will hardly address internationalisation, as
opposed to WSGUI form components.
Departing from the initial focus on web services,
WSGUI concepts are already used in other areas such
as configuration dialogs and printer dialogs, which
are generated automatically from the underlying con-
figuration schema (Schumacher, 2005). The num-
ber of shared concepts is high enough to warrant a
broader scope of WSGUI, separating the web service-
specific concepts from the generic ones. The former
are centred around web service descriptions such as
the common WSDL (Web Service Description Lan-
guage), while the latter are focusing on the generation
of forms from the message, its parts and their data
types.
6 CONCLUSION
Interfaces between web services and humans are
clearly on the agenda of researchers and influence fu-
ture service development. The WSGUI concepts col-
lectively presented in this paper, while not completely
eliminating the need for custom interfaces to web ser-
vices, are powerful enough to make the dynamic in-
vocation of web services feasible. The conditional us-
age of implicit and explicit GUI hints are of great help
in separating the work of web service developers and
interface designers.
The Dynvoker implementation handles complex
datatypes and namespaces within the XML Schema
part of WSDL files, unlike many other implementa-
tions which are only suitable for relatively simple ser-
vices. Its output in the XForms format ensures type-
safe data submission. Finally, the rigorous use of WS-
GUI concepts including several best-effort strategies
when no implicit or explicit GUI generation hints are
provided lead to usable GUIs without much effort.
Implicit semantic-driven GUI hints will be one of the
main topics of research in the near future.
REFERENCES
Bleyh, N. (2006). Analyse und Vergleich von Ans
¨
atzen zur
Einbindung von menschlichen Interaktionen in kom-
plexe Web Services. Master’s thesis, TU Dresden.
Boyer, J. M., Landwehr, D., Merrick, R., Raman, T., et al.
(2006). XForms 1.0 (W3C Recommendation), 2nd
edition.
http://www.w3.org/TR/xforms/
.
Kassoff, M., Kato, D., and Mohsin, W. (2003). Creating
GUIs for Web Services. IEEE Internet Computing,
7(4):66–73.
Kassoff, M., Spillner, J., et al. (2006). GUIDD: GUI De-
ployment Descriptor file format specification, 0.99.5
edition.
Kelly, K. E. et al. (2006). XML Forms Generator. IBM
alphaWorks.
Khushraj, D. and Lassila, O. (2005). Ontological approach
to generating personalized user interfaces for web ser-
vices. Lecture Notes in Computer Science, 3729:916–
927.
Kruschinski, V. (1999). Layoutgestaltung grafischer Be-
nutzungsoberfl
¨
achen. PhD thesis, Ruhruniversit
¨
at
Bochum.
Moebius, J. (2006). xsdtransformer.
http://xsdtrans.
sourceforge.net/#xsd2fx
.
Schumacher, C. (2005). KXforms - automatic configuration
dialog generation. aKademy conference.
Spillner, J., Bleyh, N., et al. (2006). Project Dynvocation.
http://dynvocation.selfip.net/
.
Steele, R., Khankan, K., and Dillon, T. (2005). Mobile
web service discovery and invocation through auto-
generation of abstract multimodal interface. itcc,
2:35–41.
FLEXIBLE HUMAN SERVICE INTERFACES
85