A Semantic User Interface for Managing Knowledge in Software Development
Dimitris Panagiotou and Gregoris Mentzas
School of Electrical and Computer Engineering, National Technical University of Athens
9 Iroon Polytechniou Street, Athens, Greece
Keywords: Semantic User Interface, Software Development, Knowledge Workbench, Knowledge Management in
Software Development, KnowBench.
Abstract: Modern software development consists of typical knowledge intensive tasks, in the sense that it requires that
software developers create and share new knowledge during their daily work. In this paper we propose
KnowBench a knowledge management system integrated inside Eclipse IDE that supports developers
during the software development process to produce better quality software. The goal of KnowBench is to
support the whole knowledge management process when developers design and implement software by
supporting identification, acquisition, development, distribution, preservation, and use of knowledge – the
building blocks of a knowledge management system.
Modern software development consists of typical
knowledge intensive tasks, in the sense that it
requires that software developers create and share
new knowledge during their daily work. Since
software development is a highly collaborative task,
developers are in need of simple and easy-to-use
tools that also enable collaborative work.
Knowledge has to be collected, organized,
stored, and easily retrieved when it needs to be
applied. Probst (1997) has well described the
building blocks of KM systems: identification,
acquisition, development, distribution, preservation,
and use of knowledge. Many knowledge problems
occur because organizations neglect one or more of
these building blocks.
In this paper we propose KnowBench
(knowledge workbench) which strives to provide an
intelligent, semantic user interface environment – a
knowledge management system integrated inside a
widely used software development environment
namely Eclipse IDE ( that
supports developers during the software
development process to produce better quality
software. The goal of KnowBench is to collect
knowledge the developers gain during software
development in order to avoid mistakes and leverage
successes in future projects.
KnowBench mainly supports software
developers in resolving error handling and
component reuse problems by providing an
ontology-based knowledge management system that
is build upon the KM systems’ building blocks.
The rest of the paper is organized as follows. In
the next section, we depict KnowBench’s
architecture and core components. Thereafter, we
give the results of its evaluation which draws the
guidelines for further optimization and
improvements of the system. Finally, we conclude
the paper.
2 KnowBench
The KnowBench system can be used to capture the
knowledge and experience generated during the
software development process. Although every
software development project is unique in some
sense, sharing similar experiences can assist
developers to perform their activities in a better way.
Moreover, reusing knowledge can prevent the
repetition of past failures and guide the solution of
recurrent problems.
KnowBench provides functionality that can be
used for articulating and visualizing formal
descriptions of software development related
knowledge in a flexible and lightweight manner.
Panagiotou D. and Mentzas G. (2010).
KNOWBENCH - A Semantic User Interface for Managing Knowledge in Software Development.
In Proceedings of the 5th International Conference on Software and Data Technologies, pages 508-511
DOI: 10.5220/0003006105080511
This knowledge is then retrieved and used in a
productive manner by a semantic search engine and
a P2P metadata infrastructure – namely GridVine
(Mauroux, 2007). Thus, the collaboration of
dispersed software developers is achieved who can
benefit from each others’ knowledge about specific
problems or the way to use specific source code
while developing software systems.
2.1 Ontologies
We have deployed software development ontologies
(Georgousopoulos, 2007) inside the KnowBench
system in order to describe and capture knowledge
related to software artefacts. The system architecture
allows for the extension or the use of different
software development ontologies. The set of the
deployed ontologies includes three main ontologies
(artefact, problem/solution and annotation) which
interact among each other and are all under a
common parent ontology (KnowBench ontology)
which binds them.
The artefact ontology describes different types of
knowledge artefacts such as the structure of the
project source code, reusable components, software
documentation, knowledge already existing in some
tools, etc. The problem/solution ontology models the
problems occurring during the software development
process as well as how these problems can be
resolved. The annotation ontology describes general
software development terminology as well as
domain specific knowledge. This ontology provides
a unified vocabulary that ensures unambiguous
communication within a heterogeneous community.
This vocabulary can be used for the annotation of
the artefacts.
2.2 Architecture
Figure 1 depicts the overall KnowBench
architecture. The core components delivering
KnowBench’s functionality are: semantic search,
global metadata store, software development
semantic wiki (DevWiki) and (manual/semi-
automatic) semantic annotation of source code. In
the rest of this section we give more details about
their actual interaction.
The above components are based on these APIs:
semantic annotation API, shallow NLP and IE
(information extraction) API (wrapping up and
customizing Text2Onto (Cimiano, 2004)), the
semantic search API and the global metadata store.
The global metadata store API provides an
abstract layer on top of the local and P2P
repositories through customized APIs (Jena and
GridVine-based APIs (McBride, 2002, Mauroux,
Figure 1: KnowBench architecture.
2.3 Core Components
Below we outline the core components which
constitute the KnowBench system.
2.3.1 Semantic Search
KnowBench supports advanced methods for
knowledge search through a semantic search engine
(Giesbrecht, 2008) by taking into account three
different types of search, namely keyword,
structured and semantic search.
2.3.2 Global Metadata Store
The global metadata store consists of two
components (APIs) – the LocalMDS and P2PMDS
and two managers – KeyStore Manager and Policy
Manager. It provides an abstract layer for handling
these and its purpose is to manage knowledge stored
either locally or in the P2P network.
The P2PMDS is based on the GridVine/P-Grid
system (Aberer, 2005) and is customized for
KnowBench in order to realize an access control
aware P2P metadata store using the services of
KeyStore and Policy managers.
2.3.3 Software Development Semantic Wiki
KnowBench utilizes the DevWiki system
(Panagiotou, 2009) in order to assist software
developers in the articulation and navigation of
software development related knowledge. DevWiki
uses a lightweight and flexible editor with auto-
KNOWBENCH - A Semantic User Interface for Managing Knowledge in Software Development
completion and popup support. Browsing through
knowledge is done like surfing through a
conventional wiki using the semantic links between
different knowledge artefacts. This browser is
available inside the Eclipse IDE so that the software
developer does not have to switch to another
external browser.
2.3.4 Manual/Semi-automatic Semantic
Annotation of Source Code
An important aspect of the KnowBench is the ability
to annotate semantically source code (Panagiotou,
2008). We have extended the standard Eclipse JDT
editor to add this possibility. The software developer
is able to annotate source code with semantic
annotation tags that are available or define new tags
and extend the used annotation ontology.
The manual semantic annotation of source code
provides granularity regarding the respective source
code fragment to be annotated. This granularity level
is restricted by the underlying Eclipse platform itself
as the IJavaElement interface is exploited to map
between source code fragments and metadata.
On the other hand, the shallow natural language
processing (NLP) and information extraction (IE)
API is used to semi-automatically annotate source
code corpora. The API is built on top of the
Text2Onto API and provides customizations suitable
for source code (e.g. java keywords are ignored such
as import, package for, while, etc.).
In order to achieve this goal in KnowBench we
exploited ontology learning (Maedche, 2004) and
information extraction techniques (Cunningham,
2002). Ontology learning is needed in order to
extract ontology concepts from source code corpora
that can be used for annotating it. On the other hand,
information extraction is needed in order to
instantiate the annotation ontology with individuals
(like in manual semantic annotation). A framework
exploiting both technologies is Text2Onto (Cimiano,
2004) which we adopted and extended in order to
implement the desired functionality.
3.1 Evaluation According to the KM
Building Blocks
We have conducted detailed evaluation of
KnowBench in small groups of software developers
in the following organizations: (1) Intrasoft
International S.A. – 2 developers, (2) Linux
Industrial Association – 2 developers, (3) TXT e-
Solutions – 2 developers and (4) Thales Research &
Technology – 3 developers. See (Samiotis, 2009) for
a comprehensive report on the KnowBench
We grouped the evaluation results according to
the KM building blocks depicted in section 2. Please
note that the knowledge identification block is not
described below as the KnowBench ontologies serve
this purpose. (Samiotis, 2009) provides details on
the KnowBench support of the knowledge
identification block.
3.2 Knowledge Acquisition
According to the respondents, KnowBench achieves
a good score (73% are positive) for its support in
acquiring existing knowledge. Regarding the
supported types of knowledge sources, 85% of the
respondents were satisfied with the support; 23%
found that additional types of knowledge sources
relevant for coding should be supported. As far as
the system’s response time, although 64% of the
respondents found it quite fast – some further
optimization of the system would be useful.
3.3 Knowledge Development
All respondents found the knowledge development
support in KnowBench clear and easy to follow and
agree that the system provides support at an
adequate level (76%). The meaning of knowledge
items is understandable for 86% of the respondents.
Regarding annotations, their meaning and purpose
were clear for 86% of the respondents. 79% of the
respondents found the granularity level of source
code that can be annotated sufficient. 92% of the
respondents found that KnowBench provides
friendly and easy-to-use forms for creating
annotations and only 36% of them considered the
manual annotation as effortful activity. On the other
hand, in semi-automatic annotation, 42% of
proposed annotations were chosen, thus 82% of the
respondents were satisfied with the suggestions.
3.4 Knowledge Sharing
Knowledge sharing in KnowBench meets the
expectation of 69% of respondents. Even though all
aspects of knowledge sharing in KnowBench are
above the threshold, only the level of details to be
specified in order to share knowledge and the
ICSOFT 2010 - 5th International Conference on Software and Data Technologies
usefulness of shared information received high
marks (greater than 90%).
3.5 Knowledge Usage
The search functionality received an average score
(57%). The respondents were satisfied with the
quantity and quality of search results. As far as
quantity of search results is concerned, 91% of the
respondents found the number of results optimal.
For 73% of the respondents the list of result did not
contain any irrelevant result. As regards the quality
of search results, 62% of the respondents confirmed
that the search results satisfy their information needs
more than average.
3.6 Knowledge Preservation
The lifecycle of the knowledge items i.e. creation,
update, deletion in KnowBench seems to be
supported well (expectation of 67% of the
respondents). Modification of knowledge is not a
time consuming function for 75% of the respondents
and can be done very easily by 73% of the
In this paper we presented the KnowBench system –
an intelligent, semantic user interface environment
for software developers which is integrated in the
Eclipse IDE. Semantic web technologies provide the
driving force to better manage knowledge in
software development activities inside KnowBench.
KnowBench offers an easy to use environment to
facilitate knowledge articulation and visualization
pertinent to software development. Additionally, it
provides means to annotate manually or semi-
automatically this kind of knowledge in order to
foster easier knowledge acquisition and sharing by
exploiting a semantic search engine and a P2P
metadata infrastructure. Thus, better and more
flexible collaboration among software developers
scattered across the globe is facilitated.
This work was partly supported by the TEAM
project, which is funded by the EU-IST program
under grant FP6-35111. The authors are responsible
for the content of this publication.
Aberer, K. Datta, A. Hauswirth, M. Schmidt, R. 2005.
Indexing data-oriented overlay networks, 31st
International Conference on Very Large Databases
(VLDB), Trondheim, 30 Aug - 2 Sep, 2.
Cimiano, P., Pivk, A., Schmidt, L.T., Staab, S., 2004.
Learning taxonomic relations from heterogeneous
sources, In Proc. ECAI 2004 Ontology Learning and
Population Workshop.
Cunningham, H., Maynard, D., Bontcheva, K., Tablan, V.,
2002. GATE: A framework and graphical
development environment for robust NLP tools and
applications, In Proc. 40th Annual Meeting of the
Georgousopoulos C., Happel H-J., Khan O., Maalej W.,
Narendula R., Ntioudis S., Panagiotou D., Stojanovic
L., 2007. TEAM Project Deliverable D9: User
Requirements and Conceptual Architecture.
Giesbrecht, E., Stojanovic, L., Tran, T., 2008. TEAM
Project Deliverable D29: Second iteration prototype of
Semantic Search.
Maedche, A., Staab, S., 2004. Ontology learning, In Proc.
Staab, S., Studer, R. (eds.), Handbook on Ontologies,
pp. 173-189. Springer.
Mauroux, P.C., Agarwal, S., Aberer, K., 2007. GridVine:
An Infrastructure for Peer Information Managment,
IEEE Internet Computing, vol. 11, no. 5, pp. 36-44
McBride, B., 2002. Jena: A Semantic Web Toolkit, IEEE
Educational Activities Department, Piscataway, NJ,
USA, pp. 55-59.
Panagiotou, D., Mentzas, G., 2009. A Semantic Wiki for
Software Development. In Proc. 13th Panhellenic
Conference on Informatics, 10 - 12 September 2009,
Corfu, Greece.
Panagiotou, D., Paraskevopoulos, F., Ntioudis, S., 2008.
TEAM Project Deliverable D31: 2
prototype of Knowledge Desktop.
Probst, G.B., 1997. Practical knowledge management: a
model that works, Prism, No. second quarter, pp.17-
Samiotis, K., Stojanovic, L., 2009. TEAM Project
Deliverable D39: Summative Evaluation Report.
SETF, 2006. A Semantic Web Primer for Object-Oriented
Software Developers,
KNOWBENCH - A Semantic User Interface for Managing Knowledge in Software Development