Descriptive data mining only brings its fruits when the results are provided to the end user in a palatable form. The vehicle for end-user delivery of mining results (and associated information such as data schema, task settings, and domain background knowledge) are so-called analytical reports. In order to manage a huge number of reports referring to different mining sessions, we designed a data mining web portal based on a content management system, together called SEWEBAR-CMS. 1 One of the requirements on the CMS was the ability to interact with semantic knowledge sources and other structured data, see [1].

The data analyst who authors an analytical report in the CMS has different possibilities of (semi-)automatically entering structured data into the text.

First, for locally stored data such as mining task/result/data descriptions exported from mining tools in PMML (Predictive Model Mark-Up Language), a CMS plugin can pick marked segments of HTML code, produced from PMML using XSLT, and insert them into the report as indicated by the analyst.

Second, sophisticated support for remote data/knowledge has been newly added. The infrastructure for this functionality allows to persistently specify -Links to queriable resources -Template queries for these resources (which can be paramatrized by the end-user at runtime) -XSLT transformations allowing to insert the results of queries as HTML fragments, either static or dynamically updated from the resources.

Currently we experiment with queriable resources in the form of native XML database (Berkeley, queried via XQuery), which stores PMML data, and semantic knowledge bases both in the form of SPARQL endpoint and Ontopia Knowledge Suite (a Topic Maps tool, queried via a Prolog-like language called tolog). Inclusion of further types of resources such as Lucene indices is in progress.

This work has been partially supported by the CSF project no.201/08/0802, and by Grant F4/15/2010 of the University of Economics, Prague.