Usually, OLAP(On Line Analytical Processing) systems provide data visualization through a multidimensional data model according to which a data fact is viewed as a mapping from a point in a space of dimensions into one or more spaces of measures. Moreover, dimensions are organized in levels which conform a hierarchy, providing a way of defining different levels of data aggregation, a central issue in data analysis. In a relational implementation of OLAP(usually called ROLAP), we can think of facts as being stored in fact tables, while each dimension is described in a dimension table. The industry solutions were built under the assumption that data in fact tables reflect the dynamic aspect of the data warehouse, while data in dimension tables represent static information. However, if we think of the data warehouse as a materialized view of data located in multiple sources, it is usual to find situations in which the structure of these sources changes, a new source is added, or an old one dropped. Any of these changes may require updates to the structure of some dimensions. Further, as multidimensional views are designed according to requirements from end users, a redefinition of the initial requirements may cause a dimension update. In this thesis we argue that accounting for dimension updates is necessary in an OLAP tool in order to avoid constantly rebuilding dimensions from scratch. Thus, we first characterize these updates and study the view maintenance problem when they occur. We developed algorithms which, taking advantage of the nature of the dimension updates, in some cases outperform well-known view maintenance algorithms. We then propose an extension to the MDX language(a standard query language for OLAP) and describe the implementation of TSOLAP, a multidimensional repository which supports dimension updates and view maintenance, developed following the OLE DB for OLAP standard. We discuss the experimental results of tests performed over a real-life case study, a medical center in Buenos Aires. In the second part of the thesis we embed our proposal in the temporal database framework, introducing the Temporal Multidimensional Data Model, and a temporal query language for OLAP which we called TOLAP. TOLAP allows expressing complex OLAP queries in an elegant and declarative fashion. We discuss issues like syntax, semantics, safety and expressive power. We also present an implementation including a graphic environment for temporal OLAP. Finally, we show how the temporal approach can be applied to the case study mentioned above.
hola chau _gs.DocumentHeader_ chau2 _documentheader_ chau3