ref: https://nikkie-ftnext.hatenablog.com/entry/why-rouge-score-library-cannot-calculate-from-japanese-texts

rouge_japanese_patch.py

from rouge_score.rouge_scorer import RougeScorer
from rouge_score.tokenize import SPACES_RE
from rouge_score.tokenizers import Tokenizer


class NonAlphaNumericSupportTokenizer(Tokenizer):
    """
    >>> NonAlphaNumericSupportTokenizer().tokenize("いぬ ねこ")
    ['いぬ', 'ねこ']
    """

    def tokenize(self, text):
        return SPACES_RE.split(text.lower())


if __name__ == "__main__":
    scorer = RougeScorer(
        ["rouge1", "rouge2", "rougeL", "rougeLsum"],
        tokenizer=NonAlphaNumericSupportTokenizer(),
    )
    print(scorer.score("いぬ ねこ", "いぬ ねこ"))