Developing both structured and unstructured representations,
including explicit versus implicit knowledge, as well as white-box
versus black-box approaches is a richness for Knowledge Discovery (KD).
Formal Concept Analysis (FCA) is positioned at the spectrum end which
focuses on complex data, structured representations, explicit knowledge,
explainable results and white-box approaches. More and more connections
are also made with other methods in the whole KD spectrum. Relational
Concept Analysis (RCA) extends FCA by highlighting knowledge in
multi-relational datasets. It has a variety of applications, in
particular in Software Engineering and Environmental data. As a KD
method, careful and adequate data modeling and preparation are crucial
for a successful application. Experience gained during past theoretical
or experimental studies can be capitalized in modeling patterns, by
identifying and collecting different recurring situations and adopted
solutions. Some of them can be used in other KD methods. This talk
presents the principles of RCA, its specific value for KD and its
tooling. It also situates RCA among the FCA-based KD methods. Then it
introduces a typical modeling pattern ('Separate/gather viewpoints') and
outlines a catalog. The 'Separate/gather viewpoints' pattern is applied
to the problem of feature location assistance in software product lines.
By building the pattern catalog, we aim to facilitate RCA practice, save
time and increase quality in future applications, and consolidate an
efficient KD method.
Marianne Huchard is a Full Professor of Computer Science at
Montpellier University since 2004, where she teaches courses in
knowledge engineering and software engineering. She develops her
research at LIRMM (Laboratory of Informatics, Robotics, and
Microelectronics at Montpellier). She obtained a Ph.D. in Computer
Science in 1992, during which she investigated algorithmic questions
connected to the management of multiple inheritance in various
object-oriented programming languages. She is leading research work in
Formal Concept Analysis (FCA) for more than 25 years. She contributed to
various aspects of FCA: efficient algorithms; relational concept
analysis (RCA), a framework that extends FCA to multi-relational
datasets; the connection between RCA and other formalisms, such as
propositionalization or description logics; methodology and application
of FCA to several domains, including environmental datasets, ontology
engineering, and as well software engineering driven by knowledge
engineering and FCA.
Marianne Huchard, LIRMM, Université de Montpellier, France https://marianne-huchard.fr/.