Application Knowledge Engineering Methodology
AKEM
stands for Application Knowledge Engineering Methodology.
It is an knowledge engineering methodology practiced and evolved in FF POIROT.
It is a collection of strategies and heuristics in knowledge capture,
representation and application. It is devoted to the team work with members from
multiple disciplines and different geographical locations. A key principle of
its development is the ease of practice and adaptation with emphasis on low
ceremony and agility, considering the features of knowledge engineering and its
dynamic contexts.
Life cycle model
AKEM
development life cycle model inherits from the practical experience in software
engineering and expert
system development, Rational Unified Process in
particular. It organises knowledge engineering projects
through four phases: inception, elaboration, construction and transition. Each
phase is one or more iteration of 8
activities
with different degrees of emphasis and intensity: Problem Determination,
Scoping, Analysis, Development, Deployment, Test and Validation, Documentation,
Control. They are composed of 23 tasks with 21 specific deliverables and
recommendations for implementation (17 principles, 50 best practices to follow,
11 prompt questions and 7 pitfalls). It recognises the importance of knowledge
management in knowledge engineering and stresses the facility to trace back to
the scope specification and conceptual context in order to recapture or
re-examine the previous modelling decisions and builds the traceability into
deliverables in AKEM to enable links among stories, knowledge analysis, ontology
and deployment specification.
|
Activities |
Tasks |
Deliverables |
|
Problem Determination |
Problem definition |
Vision statement |
|
Solution prescription |
Objective specification |
|
|
Scoping |
Scoping problem space |
Knowledge resources |
|
Scoping semantics |
Story |
|
|
Scoping tasks |
Task specification |
|
|
Analysis |
Knowledge constituency analysis |
Knowledge breakdown |
|
Knowledge elaboration |
||
|
Task hierarchy analysis |
Task hierarchy |
|
|
Development |
Extraction |
Highlights |
|
Paraphrases |
||
|
Abstraction |
Lexons |
|
|
Organization |
Architecture of ontology |
|
|
Groups of lexons |
||
|
Deployment |
Rule specification |
Commitments to lexons |
|
Test & Validate |
Unit validation |
Validated stories, task specification, knowledge constituency, lexon groups, commitments |
|
Integrated validation |
Validated scope, analysis, development and deployment |
|
|
Unit test |
Tested tasks, commitments |
|
|
System test |
Tested task hierarchy |
|
|
User test |
Tested stories |
|
|
Documentation |
Unit traceability |
Links across versions and deliverables |
|
System documentation |
System architecture, functionality description |
|
|
Control |
Feasibility and risk management |
|
|
Life cycle management |
||
|
Work flow management |
||
|
Quality management |
||
|
Communication management |
Scoping knowledge
Scoping
on the semantic space is less tangible than on information system
functionalities. It is not only to identify the part of the semantics proper to
focus on, but also to convey its domain contexts. The scoping activity in the
domain perspective in AKEM produces two main deliverables: knowledge resources
(documents, interview proto-cols) and stories (knowledge use cases). The
semantic scope under consideration at a given time of knowledge engineering is
specified and documented by a story. It not only identifies the focus or
boundary of attention, but also conveys the semantic context in which it stands.
The Settings of the story describe the background information. The Characters
the actors or objects involved. The Episodes describe
either sets of objects or relationships in hierarchy or a sequence of events.

Analysing knowledge
The analysis activity produces the knowledge constituent model and task hierarchy. The knowledge constituent model consists of the knowledge breakdown and the elaboration of each constituent. The knowledge breakdown seeks to modularise knowledge in a hierarchical structure and the knowledge elaboration provides the description of each constituent in a programme specification language to capture the concerned business logic.

Ontology development
The
knowledge artefact to develop is not only a of business logic and rules but also
the underlying meta-knowledge in the form of lexons (context-term-role
relations). The application specific constraints and rules are the special
commitments to the lexons The purpose is to maximise the reusability and
versatility of knowledge resources over different applications, time and
versions. Ontology is extracted from knowledge resources, such as
regulations, requirements specification, abstracted into term-role tuples
and organized into an architecture reflecting the knowledge structure of
the expert of the subject.
AKEM adopts an approach of machine-assisted human knowledge engineering. It is proven that in FF POIROT, the automatic knowledge discovery can be effectively used to perform the task of extraction. With the machine results, the task of abstraction and organization can be facilitated, especially when the coverage of the domain is large and complex.
Ontology deployment
The deployment of ontology considers system specific, application specific features to constrain lexons to produce commitments. The deployment consists of two tasks: the specification of the ontological commitments in the light of specific processing tasks and transformation of the commitments into a particular knowledge specification such as OWL, XTM, KIF.
FF POIROT experience
The ontology engineering activities in FF POIROT was planned and controlled following AKEM. Several iterations of major deliverables and small experiments were performed to augment its principles and best practices.
References
G. Zhao, J. Kingston, K. Kerremans, F. Coppens, R. Verlinden, R. Temmerman, R. Meersman, Engineering an Ontology of Financial Securities Fraud, Workshop of Regulatory Ontology, OTM 2004.
K. Kerremans, R. Temmerman, G. Zhao, Terminology and Knowledge Engineering in Fraud Detection, International conference on Terminology and Knowledge Engineering, Copenhagen, 16th - 19th August 2005.
I. Jacobson, G. Booch and J. Rumbaugh, The Unified Software Development Process, Boston, Addison-Wesley, 1999.
Kroll, P., Kruchten, P.: The Rational Unified Process Made Easy: a Practitioner’s Guide to the RUP, Boston, Addison-Wesley (2003)