Semantic Knowledge Representations for Soft Data Fusion - Efficient Decision Support Systems: Practice and Challenges from Current to Future - page 3

0

1

Semantic Knowledge Representations
for Soft Data Fusion

Claire Laudy
Thales
France

1. Introduction
Within the fusion community, studies have long been focused on the issue of fusing data that
comes from physical sensors. This type of data is also called “hard data” as the data items
represent physical measurements. Recently the issue of fusing information called “soft data”
has come to the fore. Soft data is generated by humans and may be expressed as natural
language or semi structured data. Soft data processing is a major issue for many support
systems. The new intelligence support systems for instance, aim at taking the advantage of all
types of data, and among them soft data available on the World Wide Web.
The aim of this chapter is to present the issue of soft data fusion and focus on one possible
approach that allows taking into account the discrepancies that may be observed among
different pieces of data.
The chapter is organized as follows. Section two gives an overview of what “soft data” is, as
opposed to what is commonly named “data” or “hard data” within the fusion community.
Relying on existing studies, we give an overview of soft data characteristics. We also
emphasise on the need that has recently been identify, to take into account soft data within
decision support systems.
Section three aims at giving the status and context of soft data fusion within the wide range
of fusion approaches and studies. In order to explain the context of soft data fusion first, we
present some of the different models that exist that aim at classifying fusion systems. Then, we
focus on studies related to the fusion of graph structures, as they appear to be key structures
for soft data representation. Given the exposed context, we then describe our proposition of
framework for soft data fusion which is based on three main phases: the modeling of the
application domain, an association phase and finally a multi-source fusion phase.
As we will see, soft data encompasses a high level of semantic. Therefore, semantic
representation formalisms are needed for soft data representation and fusion. Section four is
dedicated to the description of several semantic representation formalisms such as semantic
nets, ontologies and conceptual graphs.
The fifth section is dedicated to the detailed description of our proposition for soft data
fusion. The three phases defined before are detailed with a proposition of approach. We
first describe a case study that will be used in order to illustrate our approach. It concerns
TV program description fusion, within the context of a smart TV program recommendation
system. Conceptual graphs are used for domain modeling. Therefore, we will focus on
this semantic representation formalism and explain how we use it. We then describe the