Online Analytical Processing
Collected and integrated data in a data warehouse can be the subject of many interesting analyses. OLAP tools support decision-makers in analytical activities such as calculations and modeling, trend testing, and sharing of cross-sectional data. They provide answers to such questions as: who?, what?, when?, what if?, and why?. For example, in the field of customer analysis, OLAP tools enable multidimensional segmentation, determining the value of customers during their purchasing activity as well as identifying the most profitable and non-profit customers, determining customer loyalty, identifying the shopping basket for specific customer groups, and testing customer response to advertising campaigns.
The roots of OLAP tools are derived from databases and advanced mathematical models. Conceptually, the OLAP model can be presented as a hypercube that has measures inside, while dimensions are its edges. All data analyses most often relate to different levels of detail, which is why the dimensions have an internal structure that facilitates the transition from general to detail. Dimensions in the OLAP model can describe key aspects of organization’s operation, such as time, product, service, customer, distribution channels, personnel, fixed assets, etc. On the other hand, measures are usually numerical values that are the subject of analysis such as sales, number of items, number of failures, number of complaints, etc. Operations supporting multidimensional data analysis usually include mining down, mining up, mining across, and mining around.
The analysis of the IT market shows that various specialized tools are used to create multidimensional analyses, e.g., SAS OLAP Cube Studio, Oracle Warehouse Builder, and Microsoft SQL Server 2000 with Analysis Services.
Data Extraction
While a data warehouse with multidimensional analysis allows for aggregated data to be processed and grouped according to various analysis criteria, data extraction goes a step further and allows discovering new knowledge. Discovered knowledge can be valuable in predicting future events, e.g., customer behavior or opportunities to enter new markets. Such knowledge can be represented in the form of rules, correlations, exceptions, and trends and can mainly be used to describe and forecast the future.
Data extraction techniques allow discoveries of relationships that cannot be found using traditional data analysis methods. Generally speaking, data extraction means discovering various types of patterns, generalizations, regularities, properties, rules in the available data resources, and thus something that constitutes new knowledge. Gartner Group defines data extraction as a process of discovering important relationships (correlations), patterns, and trends by searching large amounts of data stored in data repositories. Nowadays, data extraction uses achievements in various fields, in particular statistics, databases and data warehouses, machine learning, data visualization, and decision making.
Knowledge resulting from data extraction can be used to predict or describe reality. Forecasting involves using known variable values to predict the future. For example, the forecasting model helps, based on data from previous periods, estimate revenues within individual product groups or customer groups. In turn, the description of reality with the help of data extraction techniques consists in creating a human readable and understandable representation of knowledge extracted from the data in the form of charts, formulas, rules, and tables. Generated knowledge, e.g., on customer purchases, can be used to support price policy decisions (Moss &C Alert, 2003).
Data extraction techniques of particular interest and usefulness in the decisionmaking process include:
■ association analysis, which consists in recognizing certain events and processes and relating them. Information about association rules is used in making decisions about preparing promotions for specific customer groups or product shelf-stacking;
■ time dependency analysis, identifying time relationships between events. This type of data mining raises high interest, inter alia, among telecommunications service operators (e.g., discovering knowledge about telecommunications alerts);
■ classification (the most widely known technique of data extraction), consisting in assigning objects (with specific attributes) to predefined classes. The classification technique can be used to skillfully conduct advertising campaigns, identify business fraud, and determine the level of risk;
■ cluster analysis, involving the allocation of objects to specific groups in such a way that the objects in one group are more similar to each other than to any object assigned to another group. Such algorithms are suitable for use in customer segmentation, product grouping, etc. Another method is the discovery of exceptions and anomalies (e.g., particularly demanding customers, expensive products); and
■ forecasting, the purpose of which is to predict the value of output attributes based on known input attribute values. Forecasting usually uses historical, statistical, and repeatable data. Prediction techniques are used, among others, for forecasting traffic in telecommunications networks, sales volume, and employee turnover.
When discussing the issue of data extraction, it should be noted that it is increasingly related to online resources. This type of data extraction is called Web mining. Web mining can mean (Kantardzic, 2002; Thuraisingham, 2003)
■ discovering new information and knowledge based on data contained in documents and directly on websites (Web Content Mining) (Rasmussen, Goldy, & Solli, 2002; Wells & Hess, 2004;);
■ discovering the structure of websites and links to other websites (Web Structure Mining) (Poul & Gautman, & Balint, 2003; Perkowitz & Etzioni, 1999); and
■ discovering user behaviors and interactions on websites (Web Usage Mining).
Visualization and Dashboards
Even the most advanced data analysis may be considered of little use if its results are not properly presented to the user. Dashboards are tools for data visualization in the form of graphic indicators, charts, and tables. They are most often available through web browsers, which means that the way to reach data is intuitive and does not require specialized IT knowledge. Dashboards are often treated as a type of interface designed to provide users with information, e.g., on firm’s condition and market trends. Based on key KPIs and relevant reports, managers can better assess and understand the firm’s situation.
