matcher.py

import string
import nltk
from nltk.translate.bleu_score import sentence_bleu
from nltk.corpus import stopwords
from nltk.stem import WordNetLemmatizer

nltk.download('wordnet', quiet=True)
lemmatizer = WordNetLemmatizer() 

class Matcher:
    @staticmethod
    def bowMatch(ref, ex, ignoreStopwords, ignoreCase):
        s1 = ref.bow()
        s2 = ex.bow()
        if ignoreCase:
            s1 = s1.lower()
            s2 = s2.lower()
        
        s1Words = s1.split(' ')
        s2Words = s2.split(' ')
        
        if ignoreStopwords:
            s1Words = Matcher.removeStopwords(s1Words)
            s2Words = Matcher.removeStopwords(s2Words)
            
        return sorted(s1Words) == sorted(s2Words)
    
    @staticmethod
    def bleuMatch(ref, ex, ignoreStopwords, ignoreCase):
        sRef = ref.bow()
        sEx = ex.bow()
        bleu = sentence_bleu(references = [sRef.split(' ')], hypothesis = sEx.split(' '))
        return bleu > Matcher.BLEU_THRESHOLD
    
    @staticmethod
    def lexicalMatch(ref, ex, ignoreStopwords, ignoreCase):
        sRef = ref.bow().split(' ')
        sEx = ex.bow().split(' ')
        count = 0
        gold_predicate_words = [lemmatizer.lemmatize(w) for w in ref.elementToStr(ref.pred.strip()).split(' ')]
        ex_predicate_words = [lemmatizer.lemmatize(w) for w in ex.elementToStr(ex.pred.strip()).split(' ')]

        # Make sure the predicates, at the very least, match before 
        # seeing if the doing the lexical match
        if not bool(set(ex_predicate_words) & set(gold_predicate_words)):
            return False

        for w1 in sRef:
            for w2 in sEx:
                if w1 == w2:
                    count += 1
                    
        # We check how well does the extraction lexically cover the reference
        # Note: this is somewhat lenient as it doesn't penalize the extraction for
        #       being too long
        coverage = float(count) / len(sRef)

        
        return coverage > Matcher.LEXICAL_THRESHOLD
    
    @staticmethod
    def removeStopwords(ls):
        return [w for w in ls if w.lower() not in Matcher.stopwords]
    
    # CONSTANTS
    BLEU_THRESHOLD = 0.4
    LEXICAL_THRESHOLD = 0.25 # Note: changing this value didn't change the ordering of the tested systems
    stopwords = stopwords.words('english') + list(string.punctuation)