Amir Netz

Integrating data mining with SQL databases: OLE DB for data mining

The integration of data mining with traditional database systems is key to making it convenient, easy to deploy in real applications, and to growing its user base. We describe the new API for data mining proposed by Microsoft as extensions to the OLE DB standard. We illustrate the basic notions that motivated the APIs design and describe the key components of an OLE DB for the data mining provider. We also include examples of the usage and treat the problems of data representation and integration with the SQL framework. We believe this new API will go a long way in enabling deployment of data mining in enterprise data warehouses. A reference implementation of a provider is available with the recent release of Microsoft SQL Server 2000 database system.

Data mining platform for database developers

For years data mining techniques have been used to analyze data residing in relational databases. Accordingly, various database vendors found innovative ways to create forms of integration between the relational database and the data mining process. However, for the most part the data mining process was never a natural component of the relational model and applying data mining processes still requires significant understanding of a specialized data model and terminology. In this talk a framework for a unified data model for relational and data mining technologies will be offered and discussed. The framework uses relational concepts to define data mining models, train them and perform prediction queries on the data mining models all using standard SQL statements with a few natural extensions. The strong relational orientation of the framework makes it easy for database developers to adopt the data mining process and to create integrated relational-data mining applications. The framework is offered as an industry standard in the OLE DB for DM specification. The talk will be coupled with live demonstrations.