使用OWL API + Pellet获取推理解释遇阻：OWLAxiom问题求助

我尝试用OWL API结合Pellet Reasoner实现类似Protégé的推理解释功能，但在构造OWLAxiom时卡壳了，不清楚该如何定义要解释的推理结论，也不确定调用的方法是否正确。

附代码如下：

import java.io.IOException;
import java.io.PrintWriter;
import java.util.Set;
import org.semanticweb.owlapi.model.*;
import com.clarkparsia.owlapi.explanation.PelletExplanation;
import com.clarkparsia.owlapi.explanation.io.manchester.ManchesterSyntaxExplanationRenderer;
import com.clarkparsia.owlapiv3.OWL;
import com.clarkparsia.pellet.owlapiv3.PelletReasoner;
import com.clarkparsia.pellet.owlapiv3.PelletReasonerFactory;

public class Main {

    private static final String file = "file:/folder/ontology.owl";
    private static final String NS = "http://www.semanticweb.org/oliver/ontologies/ontology.owl#";

    public static void main(String[] args) throws OWLOntologyCreationException, OWLException, IOException {
        PelletExplanation.setup();

        ManchesterSyntaxExplanationRenderer renderer = new ManchesterSyntaxExplanationRenderer();
        PrintWriter out = new PrintWriter(System.out);
        renderer.startRendering(out);

        OWLOntologyManager manager = OWL.manager;
        OWLOntology ontology = manager.loadOntology(IRI.create(file));

        PelletReasoner reasoner = PelletReasonerFactory.getInstance().createReasoner(ontology);

        PelletExplanation expGen = new PelletExplanation(reasoner);

        OWLIndividual angle = OWL.Individual(NS + "angle_C");
        OWLDataProperty hasAngle = OWL.DataProperty(NS + "has_angle_of");

        OWLAxiom axiom; // 不知道这里该填什么

        Set<Set<OWLAxiom>> explanations = expGen.getEntailmentExplanations(axiom); // .getEntailmentExplanations是正确的方法吗？
        out.println("Why " + angle + " " + hasAngle + " - ");

        renderer.render(explanations);
        renderer.endRendering();
    }
}

解决方法 #

1. 构造要解释的目标公理

OWLAxiom需要对应你想解释的具体推理结论，比如数据属性断言、类归属断言等，根据需求构造对应的公理类型：

• 如果要解释angle_C通过has_angle_of关联某个数值（比如90）的推理结论：

  // 构造推理出的数值字面量
  OWLLiteral angleValue = OWL.Literal(90);
  // 构造数据属性断言公理
  OWLAxiom axiom = OWL.DataPropertyAssertion(hasAngle, angle, angleValue);
  

• 如果要解释angle_C属于某个类（比如Angle类）的推理结论：

  OWLClass angleClass = OWL.Class(NS + "Angle");
  // 构造类断言公理
  OWLAxiom axiom = OWL.ClassAssertion(angleClass, angle);
  

2. 确认方法正确性

getEntailmentExplanations(axiom)是正确的方法，它会返回推导该公理所需的最小公理集合（即解释），每个子集合都是一组能推导出目标公理的本体公理。

3. 完整修改后的代码示例

import java.io.IOException;
import java.io.PrintWriter;
import java.util.Set;
import org.semanticweb.owlapi.model.*;
import com.clarkparsia.owlapi.explanation.PelletExplanation;
import com.clarkparsia.owlapi.explanation.io.manchester.ManchesterSyntaxExplanationRenderer;
import com.clarkparsia.owlapiv3.OWL;
import com.clarkparsia.pellet.owlapiv3.PelletReasoner;
import com.clarkparsia.pellet.owlapiv3.PelletReasonerFactory;

public class Main {

    private static final String file = "file:/folder/ontology.owl";
    private static final String NS = "http://www.semanticweb.org/oliver/ontologies/ontology.owl#";

    public static void main(String[] args) throws OWLOntologyCreationException, OWLException, IOException {
        PelletExplanation.setup();

        ManchesterSyntaxExplanationRenderer renderer = new ManchesterSyntaxExplanationRenderer();
        PrintWriter out = new PrintWriter(System.out);
        renderer.startRendering(out);

        OWLOntologyManager manager = OWL.manager;
        OWLOntology ontology = manager.loadOntology(IRI.create(file));

        PelletReasoner reasoner = PelletReasonerFactory.getInstance().createReasoner(ontology);
        // 预计算推理，确保所有结论都已推导完成
        reasoner.precomputeInferences();

        PelletExplanation expGen = new PelletExplanation(reasoner);

        OWLIndividual angle = OWL.Individual(NS + "angle_C");
        OWLDataProperty hasAngle = OWL.DataProperty(NS + "has_angle_of");

        // 构造要解释的目标公理：这里以数据属性断言为例
        OWLLiteral angleValue = OWL.Literal(90);
        OWLAxiom axiom = OWL.DataPropertyAssertion(hasAngle, angle, angleValue);

        Set<Set<OWLAxiom>> explanations = expGen.getEntailmentExplanations(axiom);
        out.println("Why " + angle + " has value " + angleValue + " via " + hasAngle + ":");

        renderer.render(explanations);
        renderer.endRendering();
    }
}

注意事项

• 确保目标公理确实是本体通过推理能得出的结论，否则explanations会是空集合。
• 调用reasoner.precomputeInferences()可以确保推理器完成所有推理计算，避免遗漏结论。

内容的提问来源于stack exchange，提问作者Oliver Smith