Article Preview
Top1. Introduction
Decision-support systems are tools that make it possible to gain useful information from data. In particular, Business Intelligence (BI) systems, such as OLAP and data mining, represent a particular kind of DSS that allows decision-makers to explore warehoused data to aggregate and reveal unknown trends. BI tools are usually applied in the marketing and agricultural domains, among others. Alert systems are DSSs that inform users about some particular trend or outlier in the data. Risks such as fire and flood as well as fraud detection are classical applications of alert systems.
Nowadays, the variety and multitude of available data and Decision Support Systems (DSSs) can make it difficult to identify the right solution for applications. Consequently, the conceptual design of Business intelligence (BI) systems appears more and more useful since, as widely demonstrated, conceptual models allow users to exclusively focus on functional requirements and put aside technological issues (Torlone, 2002). Moreover, conceptual models, when implemented in Computer-aided software engineering (CASE) tools, allow for automatic and free-error implementations, which can be translated into important time and economic gains. Motivated by the relevance of the conceptual design of BI applications (Bimonte et al., 2016), in this work, we focus on Data warehouses (DWs) and stream data. Data warehouse and On-Line Analytical Processing (OLAP) tools are technologies intended to support decision making (Kimball, 1996). Data Warehouses are collections of historical data that naturally evolve over time. Several works address multidimensional data and structures changing over time from the theoretical and implementation points of view (Golfarelli & Rizzi, 2009). Among temporal data, we can distinguish stream data. Stream data have been defined as a real-time continuous ordered sequence of items (Golab & Özsu, 2003). Thus, it is impossible to control the order of arrival of items and store the data stream in its entirety. Queries on data streams run continuously over a period of time and incrementally return results as new data arrive. In the context of Big Data, more and more stream data sources are available, such as web clicks, sensors networks, and mobile phones. These systems produce a huge quantity of data in a real-time and continuous way. Data streams require particular data models, query languages and tools (Data Stream Management System - DSMS) (Esper, 2017; Golab & Özsu, 2003) for their storage and analysis. The major difference between Database Management Systems (DBMSs) and DSMSs is that the latter allow queries on stored data and also on data streamed from several sources. In addition, queries and results are triggered when new data arrive. As data come and go in DSMSs, new kinds of queries are defined using the concept of the window (Golab & Özsu, 2003), which defines a buffer based on time or number of received items. Windows can also slide over these parameters (sliding windows). Alert systems are usually implemented using DSMSs for the previously described features.
Warehousing and OLAPing stream data (Stream Data Warehouse) have been broadly investigated in the academic and industrial communities (Golab et al., 2009; Quartet, 2017). Indeed, Stream Data Warehouses (SDWs) have been effectively used in several application domains such as social network analysis and fraud detection. Some works investigate the definition of OLAP operations on stream data (Cuzzocrea & Chakravarthy, 2010). These works investigate the feasibility of the aggregation computation on data streams by providing new query languages, indexing methods, compression techniques, and system architectures. However, despite the relevance of conceptual design, to the best of our knowledge, there is no work investigating the conceptual modeling of stream data warehouses.