Search code examples
javasemanticslexical-analysisdkpro-coresemantic-comparison

Use dkpro semantic similarity with uby


I want to compute similarity between strings with dkpro similarity (https://dkpro.github.io/dkpro-similarity/), it works, like so:

import org.dkpro.similarity.algorithms.api.SimilarityException;
import org.dkpro.similarity.algorithms.api.TextSimilarityMeasure;
import org.dkpro.similarity.algorithms.lsr.LexSemResourceComparator;
import org.dkpro.similarity.algorithms.lsr.gloss.GlossOverlapComparator;
import org.dkpro.similarity.algorithms.lsr.path.JiangConrathComparator;
import org.dkpro.similarity.algorithms.lsr.path.LeacockChodorowComparator;
import org.dkpro.similarity.algorithms.lsr.path.LinComparator;
import org.dkpro.similarity.algorithms.lsr.path.ResnikComparator;
import org.dkpro.similarity.algorithms.lsr.path.WuPalmerComparator;

import de.tudarmstadt.ukp.dkpro.lexsemresource.LexicalSemanticResource;
import de.tudarmstadt.ukp.dkpro.lexsemresource.core.ResourceFactory;
import de.tudarmstadt.ukp.dkpro.lexsemresource.exception.LexicalSemanticResourceException;
import de.tudarmstadt.ukp.dkpro.lexsemresource.exception.ResourceLoaderException;
import learninggoals.analysis.controller.settingtypes.SimilarityAlgorithm;



public class SemResourceComparator implements WordsComparator{
private LexicalSemanticResource resource;
private LexSemResourceComparator comparator;

//en lang
public SemResourceComparator(String resourcetype, SimilarityAlgorithm algorithm, String lang) {

        try {
            resource = ResourceFactory.getInstance().get(resourcetype, lang);
        } catch (ResourceLoaderException e) {
            // TODO Auto-generated catch block
            e.printStackTrace();
        }
        try {
        switch(algorithm){
        /*case ESA://this is vector
            comparator = new GlossOverlapComparator(resource, false);
            break;*/
        case GLOSSOVERLAP:          

            comparator = new GlossOverlapComparator(resource, false);
            break;
        case JIANG_CONRATH:
            comparator = new JiangConrathComparator(resource, resource.getRoot());
            break;
        case LEACOCK_CHODOROW:
            comparator = new LeacockChodorowComparator(resource);
            break;
        case LIN:
            comparator = new LinComparator(resource, resource.getRoot());
            break;
        case RESNIK:
            comparator = new ResnikComparator(resource, resource.getRoot());
            break;
        case WUPALMER:
            comparator = new WuPalmerComparator(resource, resource.getRoot());
            break;
        default:
            break;

        }
        } catch (LexicalSemanticResourceException e) {
            // TODO Auto-generated catch block
            e.printStackTrace();
        }

}

@Override
public double compareWords(String w1, String w2) {

   try {
    return comparator.getSimilarity(resource.getEntity(w1), resource.getEntity(w2));
} catch (SimilarityException e) {
    // TODO Auto-generated catch block
    e.printStackTrace();
} catch (LexicalSemanticResourceException e) {
    // TODO Auto-generated catch block
    e.printStackTrace();
}
   return 0;
}

i use the class like this:

double intermscore = comparator.compareWords(word1, word2);

I use LexicalSemanticResource as resource for comparing, it can be wordnet, wikipedia, germanet etc. Now i noticed that all of the resources i need are in uby (https://www.ukp.tu-darmstadt.de/data/lexical-resources/uby/, https://github.com/dkpro/dkpro-uby/blob/master/de.tudarmstadt.ukp.uby.lmf.api-asl/src/main/java/de/tudarmstadt/ukp/lmf/api/Uby.java).

My question is: can i replace the resource with a resource from uby so I don't have to include a new resource again each time i need one? so instead of ResourceFactory.getInstance().get("wordnet"), i want to use the uby resource, so sth like new Uby().getLexicalResource("wordnet") - however lexicalresource from uby is not the same as LexicalSemanticResource i use now for semantic comparison. So: Instead of using e.g. LexicalSemanticResource wordnet, i want to use wordnet from uby for the comparators. Is there a way to do this?


Solution

  • There currently is no way to do this. Uby resources and LSR resources are not compatible.

    There were plans to switch, but the issue has been open for a while: https://github.com/dkpro/dkpro-similarity/issues/39