import json, itertools, pyter from nltk.translate.bleu_score import SmoothingFunction, corpus_bleu class NLPFactGenerator: def __init__(self): self.gen_fact_list = [] self.evidence_list = [] def _split_into_words(self, sentences): return list(itertools.chain(*[_.split(" ") for _ in sentences])) def _get_word_ngrams(self, n, sentences): assert len(sentences) > 0 assert n > 0 words = self._split_into_words(sentences) return self._get_ngrams(n, words) def _get_ngrams(self, n, text): ngram_set = set() text_length = len(text) max_index_ngram_start = text_length - n for i in range(max_index_ngram_start + 1): ngram_set.add(tuple(text[i:i + n])) return ngram_set def load_data(self, filename): with open(filename, "r") as infile: self.data = json.load(infile) def get_title_evidence_generated_facts(self): titles = [] evidences = [] generated_facts = [] for entry in self.data: titles.append(entry["title"]) evidences.append(entry["evidence"]) generated_facts.append(entry["generated_fact"]) return evidences, generated_facts def ter(self): ref, gen = self.get_title_evidence_generated_facts() if len(ref) == 1: total_score = pyter.ter(gen[0].split(), ref[0].split()) else: total_score = 0 for i in range(len(gen)): total_score = total_score + pyter.ter(gen[i].split(), ref[i].split()) total_score = total_score/len(gen) return total_score def bleu(self): evidence_list, gen_fact_list = self.get_title_evidence_generated_facts() ref_bleu = [] gen_bleu = [] for l in evidence_list: gen_bleu.append(l.split()) for i,l in enumerate(gen_fact_list): ref_bleu.append([l.split()]) cc = SmoothingFunction() score_bleu = corpus_bleu(ref_bleu, gen_bleu, weights=(0, 1, 0, 0), smoothing_function=cc.method4) return score_bleu def rouge_one(self,n=3): evidence_list, gen_fact_list = self.get_title_evidence_generated_facts() evaluated_ngrams = self._get_word_ngrams(n, evidence_list) reference_ngrams = self._get_word_ngrams(n, gen_fact_list) reference_count = len(reference_ngrams) evaluated_count = len(evaluated_ngrams) overlapping_ngrams = evaluated_ngrams.intersection(reference_ngrams) overlapping_count = len(overlapping_ngrams) if evaluated_count == 0: precision = 0.0 else: precision = overlapping_count / evaluated_count if reference_count == 0: recall = 0.0 else: recall = overlapping_count / reference_count f1_score = 2.0 * ((precision * recall) / (precision + recall + 1e-8)) return recall if __name__ == "__main__": fact_generator = NLPFactGenerator() fact_generator.load_data("generated_facts_xlsum.json") rouge_one_score = fact_generator.rouge_one() blue_score = fact_generator.bleu() print(blue_score) print(rouge_one_score)