Detection and Propagation of Complex Evolutions in Multiversion Data Warehouse Ines Zouari Turki, Hana Boukhriss, Faiza Ghozzi Jedidi, Rafik Bouaziz MIRACL Laboratory ISIMS, Sfax technology park, PB: 242-Sakiet Ezzit, 3021 Sfax Tunisia Ines.Zouari@isimsf.rnu.tn, Boukhriss.Hana@gmail.com, Jedidi_Faiza@yahoo.fr, Raf.Bouaziz@fsegs.rnu.tn Abstract. Data warehouse (DW) instance changes can be categorized in simple changes, e.g. insert/delete dimension member, and complex changes, e.g. merge/split/regrouping dimension members. These complex instance changes can not be automatically detected from DW sources and propagated in a multiversion DW (MVDW). In this paper we propose a semi-automatic approach to detect and manage simple and complex instance evolutions in MVDW, when analysing change events described in the change logs of data sources. This approach implicates the DW administrator to specify the evolution operation and its corresponding change events in the logs, as well as to enrich change actions that will be propagated in the MVDW. Keywords: Multiversion data warehouse, Change log, Change detection, Dimension member evolution. 1 Introduction A Data Warehouse (DW) is a special database that integrates data from multiple operational systems and external sources. The most popular architecture for DW is based on multidimensional data cubes [6], where the analysis subject (fact) is described by numerical measures and several dimensions, constituting analysis axes. A dimension consists of several levels organized in one or several hierarchies that provide a way of selecting and aggregating measure values. Each level is described by a key attribute (analysis parameter) and weak attributes (secondary information). The schema part of a DW is defined by its different components (cube, dimension, etc.). Its instance part is made of level instances called dimension member (DMem) [6]. As example, consider the cube Water_Sales inspired from the commercial activity of SONEDE (National Society of Water Exploitation and Distribution in Tunisia). This cube analyses the measure Sold_Water_Quantity relatively to the dimensions Time, Water_Connection and Water_Use. These dimensions are structured in hierarchies as follows ("←" means "rolls-up to"). Time: Year ← Quarter ← Month; Water_Use: Use_Class ← Use; Water_Connection: Regional_Direction ← District ← Exploitation_Center ← Connection. Changes in DW sources, often affect the content and structure of a DW built on