Templates - Knowledge Discovery
An Active Template is a tool to help the business user develop analytic models. Active Templates provide a guide template and resource library, and they record and document the process used to build a specific model. Taken together, these attributes enable a business user to more easily understand, use and maintain an analytic model.
The tools now available for developing analytic models focus on automating the use of various data mining algorithms, but do not address the more important issue of process or methodology. They provide no means to capture the thought process of the analytic modeler, and as a result most analytic models are delivered as "black boxes" with little or no accompanying documentation. In most cases the business users do not fully understand the assumptions and limitations of these "black boxes." This lack of understanding can prevent them from maintaining the model properly, which can cause it to eventually lose effectiveness or even be misused or its output be misinterpreted.
Active Templates guides analytic modelers through an established formal methodology, and documents their activities and thought process. The NCR Knowledge Discovery Analytic Process (KDAP) consists of the processes illustrated in the image in the left column.
Each sub-process is further made up of tasks and sub-tasks. In most cases a task leads to the execution of a data analysis algorithm. The choice of a data analysis algorithm often makes additional tasks necessary. In an Active Template each task is associated with a frame data structure. The documentation in a frame represents the analytic modelers thought process: why a task was done, how it was done, what results were expected and actual results obtained. The frame also contains a link to the executable code created as a result of the task, which allows it to be re-run at any time.
Analytic model development involves a large number of data analysis algorithms. An encyclopedia stores descriptions of these algorithms, their assumptions, situations in which they should be applied, and pitfalls in applying them.
The data used for analytic model development, the documentation in the frames, and the entries in the encyclopedia interact with one another, guiding the analytic modeler to appropriate tasks in the knowledge discovery process. An Active Template is the packaging of the data, the developed code, the frames, and the encyclopedia entries that are used for a specific analytic model.
HITC has already developed Active Templates for Customer Segmentation, Cashier Monitoring, and Promotion Optimization. Active Templates are being developed for Customer Profiling, Trend Analysis and Forecasting.