Decisional Annotations: Integrating and Preserving Decision-Makers' Expertise in Multidimensional Systems

Introduction

Multidimensional data analysis consists in manipulations through aggregations of data drawn from various transactional databases. This approach is often based on multidimensional databases (MDB). MDB schemas are composed of facts (subjects of analysis) and dimensions (axes of analysis) (Ravat et al., 2008). Decision-making consists in analysing these multidimensional data. Nevertheless, due to its numeric nature it is difficult to interpret business data. This work requires decision-makers to achieve a tedious cognitive effort, which is an immaterial capital. To take relevant decisions this required expertise is very valuable but it cannot be expressed, stored, and exploited in traditional multidimensional systems. Such an expertise can be qualified as ephemeral from the organization standpoint.

As paper annotations convey information between readers (Marshall, 1998), we argue that annotations can also support this immaterial capital for MDB. We consider an annotation as a high value-added component of MDB from the users’ standpoint. Such components can be used for a personal use to remind any information concerning the data under study, as well as for a collective use to share information that makes complex analyses easier. This collective use of annotations would serve as a basis for building an expertise memory that stores previous decisions and commentaries. Moreover as Foshay et al. (2007) state “Metadata helps data warehouse end users to understand the various types of information resources available from a data warehouse/business intelligence environment.” As a consequence, in our proposition, annotations and their contents enable end users to analyse, discuss and share knowledge in context during the decision making process.

This chapter addresses the problem of integrating the annotation concept into MDB management systems. Annotations are designed to assist decision-makers and to turn their expertise persistent and reusable.

Related works and discussion. To the best of our knowledge, integrating annotations in the MDB context has not been studied yet. The closest works are related to annotation integration in Relational DataBase Management Systems (RDBMS). First, in the DBNotes system (Bhagwat et al. 2004, 2005; Chiticariu et al., 2005; Tan, 2003) zero or several annotations are associated with a relation element. Annotations are transparently propagated along as data is being transformed (through SQL queries). This annotation system traces the origin and the flow of data. Second, the authors in (Cong et al., 2006) and (Geerts et al., 2006) specify an annotation-oriented data model for the manipulation and the search of both data and annotations. This model is based on the concept of block to annotate both a single value and a set of values. A prototype, called MONDRIAN, supports this annotation model. Third, similar to the previous systems, the works presented in (Bhatnagar et al., 2007a) and (Bhatnagar et al., 2007b) consist in annotating relational data. DBNotes and MONDRIAN use relational data to express annotations whereas this last work models annotations using eXtensible Markup Language (XML). The model allows users to cross-reference related annotations.

As conceptual structures of a MDB are semantically richer, the outlined works cannot be directly applied to our context.

• •Contrary to RDBMS where a unique data structure is used to both store and display data, in our MDB context, the storage structures are more complex and a specific display is required.

• •In the RDBMS framework, annotations are straightforwardly attached to tuples or cell values (Bhagwat et al., 2004). Due to the MDB structures, annotations must be attached to more complex data; e.g. dimension attributes are organised according to hierarchies and displayed decisional data are often computed from aggregations.

To annotate a MDB we define a specific model having the following properties: