The XML_OWL Module
For this section it is assumed that the reader is familiar with the XML and OWL module. The XML_OWL module basically combines the XML and OWL module, but incorporates more background knowledge into the data-mining process by linking XML and OWL data sources. Instead of going through each workflow step, we will just focus on those steps that are relevant for this module.
DataSelection
Create a new project named xml_owl and add the following files from the examples/trains folder to the project:
- trains.xml as input for XML
- trains_scheme.owl as SCHEME for OWL
- trains_instance.owl as INSTANCE for OWL
Workflow
Create a new experiment. The workflow configuration will appear. Choose XML_data as operation for the DataSet task. This way, we will choose nodes from the XML input file and add background knowledge from the ontology to them.
Create an instance of the workflow and start it.
DataSelection
We will use the same dataset as we did for the XML module. Your dataset should look like this:
DataLinking
This task is the crucial element of the XML_OWL module. We will add background knowledge to our selected XML nodes from the ontology. To accomplish this, we need to map XML nodes to OWL individuals with the following proceeding:
- Select "<train>, depth: 1" from XML Elements
- Map all nodes of this element to "http://helenos.ontoware.org/example/trains#Eastbound" and "http://helenos.ontoware.org/example/trains#Westbound" trains by checking the corresponding boxes
- Select the "<train>, depth: 1, object: 2" node
- Map it to the OWL individual "http://trains.example#train1" by checking the box. (Notice that as soon as you check the box, the box of the XML node is also checked, giving you an overview of the XML nodes that are already mapped.)
- Map the remaining XML elements as following:
- "<train>, depth: 1, object: 6" -> "http://trains.example#train2"
- "<train>, depth: 1, object: 10" -> "http://trains.example#train3"
- "<train>, depth: 1, object: 14" -> "http://trains.example#train4"
- "<train>, depth: 1, object: 18" -> "http://trains.example#train5"
- "<train>, depth: 1, object: 22" -> "http://trains.example#train6"
Afterwards, continue the worklow.
DataMining
Take a look at the modes. You will notice that now the modes, which were generated for the XML and OWL data, both appear.
Progol uses a technique called "mode directed inverse entailment". On the one hand the idea of the modes is to guide the search through the input data. On the other hand the input data generated by Helenos take the mapping between XML nodes and OWL individuals into account. Hence, XML and OWL are integrated for the data-mining task.
Run Progol by clicking "Run" and sending the "generalise(node/1)?" command.
Remove the "modeb(*,trains_Eastbound(+object))?" mode to guide the search for a general rule into another direction. Run Progol again.
Removing the "modeb(*,trains_next_car(+object,-object))?" mode yields the following result:
Conclusions
Consider all the results we have obtained:
- node(A) :- trains_Eastbound(A).
- node(A) :- trains_next_car(A,B), trains_shape(B,trains_shape_2).
- node(A) :- has_elem_direction(A,B), has_text(B,east).
Our selection was based on some criterions that only were available in the XML input. Incorporating background knowledge from the OWL input by mapping XML nodes and OWL individuals, the first and second result show that the background knowledge can serve to derive better rules. Without the ontological background the data-mining task would have only resulted in the third rule.
Of course, this is only a starting point. Especially the mapping process is tedious. But if you have a domain described by some XML input, and there exists an ontology about it, and you are able to come up with a proper mapping, Helenos provides a way of integration in order to assist your data-mining needs.