Bernd Neumayr

Modeling techniques for multi-level abstraction

Employing multi-level abstraction in modeling refers to representing objects at multiple levels of one or more abstraction hierarchies, mainly classification, aggregation and generalization. Multiple representation, however, leads to accidental complexity, complicating modeling and extension. Several modeling techniques, like powertypes, deep instantiation, materialization, m-objects, HERM, and the component model may be used to reduce unnecessary complexity with multilevel abstraction. This chapter compares these modeling techniques using four comparison criteria: (1) compactness (modular and redundancyfree models), (2) query flexibility (number and kind of pre-defined entry points for querying), (3) heterogeneous level-hierarchies, and (4) multiple relationship-abstractions (such as between relationship occurrence and relationship type).

Dual Deep Instantiation and Its ConceptBase Implementation

Application integration requires the consideration of instance data and schema data. Instance data in one application may be schema data for another application, which gives rise to multiple instantiation levels. Using deep instantiation, an object may be deeply characterized by representing schema data about objects several instantiation levels below. Deep instantiation still demands a clear separation of instantiation levels: the source and target objects of a relationship must be at the same instantiation level. This separation is inadequate in the context of application integration. Dual deep instantiation (DDI), on the other hand, allows for relationships that connect objects at different instantiation levels. The depth of the characterization may be specified separately for each end of the relationship. In this paper, we present and implement set-theoretic predicates and axioms for the representation of conceptual models with DDI.

Towards ontology-based OLAP: datalog-based reasoning over multidimensional ontologies

Understandability, reuse, and maintainability of analytical queries belong to the key challenges of Data Warehousing, especially in settings where a large number of business analysts work together and need to share knowledge. To tackle these challenges we propose Ontology-based OLAP where an ontology acts as superimposed conceptual layer between business analysts and multidimensional data. In Ontology-based OLAP, dimensions and facts are enriched by concept definitions capturing the semantics of relevant business terms used to define measures and to formulate analytical queries. Using traditional ontology languages, it is, however, very difficult to capture the hierarchical and multidimensional conceptualizations of business analysts. In this paper, we propose hierarchical and multidimensional ontologies to better capture these structural specificities. We define and implement the abstract structure and semantics of multidimensional ontologies as rules and constraints in Datalog with negation and represent multidimensional ontologies as Datalog facts. In addition to reasoning over multidimensional ontologies (open-world) we discuss their grounding in Data Warehouses (closed-world) as the fundament of Ontology-based OLAP.

Semantic cockpit: an ontology-driven, interactive business intelligence tool for comparative data analysis

Business analysts frequently use Cockpits or Dashboards as front ends to data warehouses for inspecting and comparing multidimensional data at various levels of detail. These tools, however, perform badly in supporting a business analyst in his or her business intelligence task of understanding and evaluating a business within its environmental context through comparative data analysis. With important business knowledge either unrepresented or represented in a form not processable by automatic reasoning, the analyst is limited in the analyses that can be formulated and she or he heavily suffers from information overload with the need to re-judge similar situations again and again, and to re-discriminate between already explained and novel relationships between data. In an ongoing research project we try to overcome these limitations by applying and extending semantic technologies, such as ontologies and business rules, for comparative data analysis. The resulting Semantic Cockpit assists and guides the business analyst due to reasoning about various kinds of knowledge, explicitly represented by machine-processable ontologies, such as organisation-internal knowledge, organisation external domain knowledge, the semantics of measures and scores, knowledge about insights gained from previous analysis, and knowledge about how to act upon unusually low or high comparison scores. This paper outlines the architecture of the Semantic Cockpit and introduces its core ideas by a sample use case.

Using domain ontologies as semantic dimensions in data warehouses

More and more transaction systems collect records that reference concepts of domain ontologies in so-called semantic attributes. This paper investigates how such semantic attributes together with their referenced domain ontology can be best exploited for data analysis in data warehouses. This gives rise to two challenges: first, exploit the rich knowledge represented in domain ontologies for targeted analysis and, second, aggregate data along the subsumption hierarchy of concepts and ensuring summarizability in the absence of predefined aggregation levels. To meet these challenges, the paper extends OLAP (OnLine Analytical Processing) by integrating concept expressions and proper level definitions over domain ontologies into OLAP operations. A prototype demonstrates the feasibility of the approach.

Multi-level Conceptual Modeling and OWL

Ontological metamodeling or multilevel-modeling refers to describing complex domains at multiple levels of abstraction, especially in domains where the borderline between individuals and classes is not clear cut. Punning in OWL2 provides decideable metamodeling support by allowing to use one symbol both as identifier of a class as well as of an individual. In conceptual modeling more powerful approaches to ontological metamodeling exist: materialization, potency-based deep instantiation, and m-objects/m-relationships. These approaches not only support to treat classes as individuals but also to describe domain concepts with members at multiple levels of abstraction. Based on a mapping from m-objects/m-relationships to OWL we show how to transfer these ideas from conceptual modeling to ontology engineering. Therefore we have to combine closed world and open world reasoning. We provide semantic-preserving mappings from m-objects and m-relationships to the decideable fragment of OWL, extended by integrity constraints, and sketch basic tool support for applying this approach.

Semantic Enrichment of OLAP Cubes: Multi-dimensional Ontologies and Their Representation in SQL and OWL

A multi-dimensional ontology (MDO) enriches an OLAP cube with concepts that represent business terms in the context of data analysis. The formal representation of the meaning of business terms facilitates the unambiguous interpretation of query results as well as the sharing of knowledge among business analysts. In contrast to traditional ontologies, an MDO captures the multi-dimensional, hierarchical world view of business analysts. In this paper, we introduce a translation of MDO concepts to SQL in order to allow for the querying of a closed-world OLAP cube. We introduce a representation in OWL in order to determine subsumption hierarchies of MDO concepts using off-the-shelf reasoners.

Ontology-Driven Business Intelligence for Comparative Data Analysis

In this tutorial, we present an ontology-driven business intelligence approach for comparative data analysis which has been developed in a joint research project, Semantic Cockpit (semCockpit), of academia, industry, and prospective users from public health insurers. In order to gain new insights into their businesses, companies perform comparative data analysis by detecting striking differences between different, yet similar, groups of data. These data groups consist of measure values which quantify real-world facts. Scores compare the measure values of different data groups. semCockpit employs techniques from knowledge-based systems, ontology engineering, and data warehousing in order to support business analysts in their analysis tasks. Concept definitions complement dimensions and facts by capturing relevant business terms which are used in the definition of measures and scores. Furthermore, domain ontologies serve as semantic dimensions and judgement rules externalize previous insights. Finally, we sketch a vision of analysis graphs and associated guidance rules to represent analysis processes.

Multi-dimensional navigation modeling using BI analysis graphs

To solve analysis tasks in business intelligence, business analysts frequently apply a step-by-step approach. Using their expert knowledge they navigate between different measures, also referred to as business ratios or key performance indicators, at different levels of detail and focus on different aspects or parts of their organization. In this paper we introduce BI Analysis Graphs to document and analyze these navigation steps. We further introduce BI Analysis Graph Templates to model and re-use recurrent navigation patterns. We describe reasoning tasks over BI Analysis Graph Templates and sketch how they are implemented in our proof-of-concept prototype. BI Analysis Graph Templates may serve as formal foundation for interactive dashboards and guided analytics in business intelligence applications.

Dual deep modeling : multi-level modeling with dual potencies and its formalization in F-Logic

An enterprise database contains a global, integrated, and consistent representation of a company’s data. Multi-level modeling facilitates the definition and maintenance of such an integrated conceptual data model in a dynamic environment of changing data requirements of diverse applications. Multi-level models transcend the traditional separation of class and object with clabjects as the central modeling primitive, which allows for a more flexible and natural representation of many real-world use cases. In deep instantiation, the number of instantiation levels of a clabject or property is indicated by a single potency. Dual deep modeling (DDM) differentiates between source potency and target potency of a property or association and supports the flexible instantiation and refinement of the property by statements connecting clabjects at different modeling levels. DDM comes with multiple generalization of clabjects, subsetting/specialization of properties, and multi-level cardinality constraints. Examples are presented using a UML-style notation for DDM together with UML class and object diagrams for the representation of two-level user views derived from the multi-level model. Syntax and semantics of DDM are formalized and implemented in F-Logic, supporting the modeler with integrity checks and rich query facilities.