Semantic Knowledge Representations for Soft Data Fusion - Efficient Decision Support Systems: Practice and Challenges from Current to Future - page 4

association phase, which is carried out relying on several similarity measures and the fusion
phase which relies on an existing operation on the conceptual graphs, the maximal join ,
extended thanks to domain knowledge.
Section six describes some of the experimentations that we conducted on TV program
description, in order to demonstrate the validity and usefulness of our approach.
2. Soft data
2.1 From data to information: an abstraction process
In the domain of Artificial Intelligence (AI), data, information and knowledge are central
notions. Many definitions exist for the words “data”, “information” and “knowledge”. In
this paragraph, we give our definitions of these three concepts. They are inspired from the
socio-economical domain. Data, information and knowledge are points in a continuum along
which an abstraction process takes place.
Abstraction is the process of highlighting some of the aspects of a thing in order to grasp its
characteristics. It is somehow a process of generalization. Abstracting an observable thing
leads to a general representation of this reality, which is often called a concept .
Data items are unprocessed and uninterpreted symbols. They are elementary descriptions of
measurable properties.
Information is what data items become once they have been interpreted and contextualized
so to become useful within a specific objective and for a specific user. Having information
is “knowing what is happening”. The information answers to questions such as “who?”,
“what?”, “where?” and “when?”
Knowledge is a combination of information with experience and judgment. It allows
reasoning among information and interpreting data in order to create new data and
information items. The knowledge answers to the question “How?”.
In the specific case of fusion, the notions of data, information and knowledge are also linked
one to another within the process of abstraction (see Figure 1). The aim of information and
data fusion is to have a representation of an external situation. This representation can
be built thanks to observations of the external situation that are acquired through sensors
and reported to fusion systems. Sensors are either low-level physical sensors, that report
about physical measurements such as temperature or speed, or human observers that report
about (some parts of) complex situations. In the first case, the physical sensors give a set
of data that must be interpreted. The human sensors, on the contrary, provide interpreted
information. Combining all the information items in order to deduce new information and
pieces of knowledge is the role of the information fusion systems.
Both data and information fusion systems use domain knowledge in order to interpret and
combine the data and information items, according to a specific aim and within a specific
context. Domain knowledge is also used in order to solve inconsistencies and discrepancies
among the data and information items.
2.2 Soft data: a new challenge for decision support systems
“Soft data” items are observations that are generated by humans. They may be expressed as
unconstrained natural language (see Sambhoos et al. (2008)), through textual data or speech
signal, but can also be made of semi constrained data items such as xml files or data bases,
which are keyed in by humans through forms for instance. As soft data is provided by