Difference between revisions of "ProgramaticLuceneQueryAccess"

Revision as of 13:49, January 2, 2008

This note will describe how to programatically

• generate phonetic lucene indicies for an ontology
• make a phonetic query that uses the lucene indicies and
• save the project to include the lucene query tab.

I will assume that you have already obtained the binary or src (http://smi-protege/repos/protege/lucene-query/trunk) for the lucene query plugin. The test code I used had the following structure:

    @SuppressWarnings("unchecked")
    public static void main(String args[]) {
        List errors = new ArrayList();
        Project project = new Project(PROJECT_FILE, errors);
        displayErrors(errors);
        OWLModel om = (OWLModel) project.getKnowledgeBase();

        indexOntology(om);
        testQuery(om);
        addTab(project);
    }

The first few lines just illustrate one of the standard ways of opening an ontology in Protege. This test code was set up to work with the pizza project but it could easily be modified to fit other ontologies. The last three lines show the code that we will explain here.

First we index the ontology. This is done by the indexOntology routine and the code looks like this:

    private static void indexOntology(KnowledgeBase kb) {
        new InstallNarrowFrameStore(kb).execute();
        new IndexOntologies(kb).execute();
    }

In this code the InstallNarrowFrameStore and IndexOntologies classes are found in the package edu.stanford.smi.protege.query from the lucene query plugin. Both classes are ProtegeJob classes and will therefore work if the knowledge base is a client knowledge base accessing a remote server.

Now that we have indexed the ontology, we can do a programatic query. This is done in the following code:

    private static void testQuery(KnowledgeBase kb) {
        Slot slot = kb.getSystemFrames().getNameSlot();
        Set<Frame> results = null;
        results = kb.executeQuery(new PhoneticQuery(slot, "sheazeypeasah"));
        if (results == null || results.size() != 1) {
            System.out.println("test of search failed");
        }
        else {
            System.out.println("Search succeeded (" + results.iterator().next().getBrowserText() + ")");
        }
    }

There are several types of queries (some of which are composed out of other queries) defined in the lucene query plugin including:

• conjuctions and disjunctions of queries,
• queries that can only return a limited set of results,
• queries that find regular expression matches for a slot value,
• queries that match slot values with the results of a nested query and
• queries that look inside a restriction for matches to an inner query.

Finally we add the lucene plugin as a tab to the project. I was not quite able to use the code suggested in Adding and removing tabs programmatically because when I retrieved the project view it was null. The code is as follows:

   @SuppressWarnings("unchecked")
    private static void addTab(Project project) {
        List errors = new ArrayList();
        
        WidgetDescriptor tabWidgetDescriptor = project.getTabWidgetDescriptor("edu.stanford.smi.protege.query.LuceneQueryPlugin");
        tabWidgetDescriptor.setVisible(true);

        Collection widgets = project.getTabWidgetDescriptors();
        project.setTabWidgetDescriptorOrder(widgets);
        
        project.save(errors);
        displayErrors(errors);
    }

Issues

The indicies are stored in a slightly funny place (protege-dir/lucene/project-name/*). We may want to choose a better approach and implement that.

This should really support other types of lucene search than phonetic. It looks like this is extremely easy to do but it has been a while since I touched this code.