Examples with COMBINED_SCORE org.molgenis.ontology.sorta.meta.OntologyTermHitMetaData.COMBINED_SCORE used on opensource projects

Example 1 with COMBINED_SCORE

use of org.molgenis.ontology.sorta.meta.OntologyTermHitMetaData.COMBINED_SCORE in project molgenis by molgenis.

the class SortaServiceImpl method findSynonymWithHighestNgramScore.

/**
 * A helper function to calculate the best NGram score from a list ontologyTerm synonyms
 */
private Entity findSynonymWithHighestNgramScore(String ontologyIri, String queryString, Entity ontologyTermEntity) {
    Iterable<Entity> entities = ontologyTermEntity.getEntities(OntologyTermMetaData.ONTOLOGY_TERM_SYNONYM);
    if (Iterables.size(entities) > 0) {
        String cleanedQueryString = removeIllegalCharWithSingleWhiteSpace(queryString);
        // Calculate the Ngram silmiarity score for all the synonyms and sort them in descending order
        List<Entity> synonymEntities = FluentIterable.from(entities).transform(ontologyTermSynonymEntity -> {
            Entity mapEntity = ontologyTermSynonymFactory.create();
            mapEntity.set(ontologyTermSynonymEntity);
            String ontologyTermSynonym = removeIllegalCharWithSingleWhiteSpace(ontologyTermSynonymEntity.getString(OntologyTermSynonymMetaData.ONTOLOGY_TERM_SYNONYM_ATTR));
            mapEntity.set(SCORE, NGramDistanceAlgorithm.stringMatching(cleanedQueryString, ontologyTermSynonym));
            return mapEntity;
        }).toSortedList((entity_1, entity_2) -> entity_2.getDouble(SCORE).compareTo(entity_1.getDouble(SCORE)));
        Entity firstMatchedSynonymEntity = Iterables.getFirst(synonymEntities, ontologyTermSynonymFactory.create());
        double topNgramScore = firstMatchedSynonymEntity.getDouble(SCORE);
        String topMatchedSynonym = firstMatchedSynonymEntity.getString(OntologyTermSynonymMetaData.ONTOLOGY_TERM_SYNONYM_ATTR);
        // else move to next synonym
        for (Entity nextMatchedSynonymEntity : Iterables.skip(synonymEntities, 1)) {
            String nextMatchedSynonym = nextMatchedSynonymEntity.getString(OntologyTermSynonymMetaData.ONTOLOGY_TERM_SYNONYM_ATTR);
            StringBuilder tempCombinedSynonym = new StringBuilder();
            tempCombinedSynonym.append(topMatchedSynonym).append(SINGLE_WHITESPACE).append(nextMatchedSynonym);
            double newScore = NGramDistanceAlgorithm.stringMatching(cleanedQueryString, removeIllegalCharWithSingleWhiteSpace(tempCombinedSynonym.toString()));
            if (newScore > topNgramScore) {
                topNgramScore = newScore;
                topMatchedSynonym = tempCombinedSynonym.toString();
            }
        }
        firstMatchedSynonymEntity.set(OntologyTermSynonymMetaData.ONTOLOGY_TERM_SYNONYM_ATTR, topMatchedSynonym);
        firstMatchedSynonymEntity.set(SCORE, topNgramScore);
        firstMatchedSynonymEntity.set(COMBINED_SCORE, topNgramScore);
        // The similarity scores are adjusted based on the inverse document frequency of the words.
        // The idea is that all the words from query string are weighted (important words occur fewer times across
        // all ontology terms than common words), the final score should be compensated for according to the word
        // // weight.
        Map<String, Double> weightedWordSimilarity = informationContentService.redistributedNGramScore(cleanedQueryString, ontologyIri);
        Set<String> synonymStemmedWords = informationContentService.createStemmedWordSet(topMatchedSynonym);
        Set<String> createStemmedWordSet = informationContentService.createStemmedWordSet(cleanedQueryString);
        createStemmedWordSet.stream().filter(originalWord -> Iterables.contains(synonymStemmedWords, originalWord) && weightedWordSimilarity.containsKey(originalWord)).forEach(word -> firstMatchedSynonymEntity.set(COMBINED_SCORE, (firstMatchedSynonymEntity.getDouble(COMBINED_SCORE) + weightedWordSimilarity.get(word))));
        return firstMatchedSynonymEntity;
    }
    return null;
}