A Semantic Data Model for Integration of Data and Knowledge

In this paper we present the Data and Knowledge Model (DK-model) that integrates the representation of data and certain types of knowledge in a semantic data model. The ER model is used as a basis for the DK-model. The main features of the DK-model are modularity of modelling, generalization/spe-cialization hierarchies, deductive capabilities in the form of so-called virtual attributes, and inheritance of attributes and knowledge rules. Virtual attributes do not have a static value, but an associated knowledge rule that states how to compute their value. These knowledge rules are speciied declaratively in a Datalog-like way. Because virtual attributes are subject to inheritance they allow for a modular, non-redundant deenition of knowledge rules. The mapping of knowledge rules, which describe virtual attributes, onto Relational Algebra is also described. This leads to an eecient, set-oriented way of solving queries. The need for semantic data models has been widely recognized, and led to the development of several important models 2,4,12]. Important issues are modularity of modelling, generaliza-tion/specialization, and property inheritance 13]. Most semantic data models are limited to the modelling of static data structures. We feel that it is important to incorporate knowledge rules in these data structures as well. In this paper we describe a semantic data model, called the Data and Knowledge Model (DK-Model), that integrates the notions of generalization/specialization, property inheritance, and virtual attributes. These are attributes that do not have a static data value, but an associated knowledge rule that describes how to compute their value. These knowledge rules are described declaratively in a Datalog-like form. First we describe some of the reeections that led us to the development of the DK-model in its present form. A lot of eeort is put into adding extra reasoning power to database systems, e.g., by coupling relational databases with Prolog (for a good overview see 3]). However, when this extra reasoning power is not in some way integrated with a powerful semantic data model it is of hardly any use. The same goes for the development of semantic data models; when the concepts introduced by these models are not supported by a powerful database system the data model looses much of its actual signiicance. Therefore, we have chosen to develop a semantic data model that adds signiicant capabilities of integrating data and knowledge, and can be implemented eeciently on top of a relational database system. We noticed that several types of …