如何避免词组拆分影响 TF-IDF 计算？

from sklearn.feature_extraction.text import tfidfvectorizer

docs = ["this_is_book", "this_is_apple"]
vectorizer = tfidfvectorizer(analyzer="word", stop_words="english")
tfidf = vectorizer.fit_transform(docs)
print(vectorizer.get_feature_names_out())

['this_is_apple', 'this_is_book']

import math

def tfidf_custom(docs, vocab):
  """
  自定义 TF-IDF 计算

  参数：
    docs: 文档集合
    vocab: 词汇表
  """

  # 计算词频
  tf_dict = {}
  for doc in docs:
    for word in doc:
      if word in tf_dict.keys():
        tf_dict[word] += 1
      else:
        tf_dict[word] = 1

  # 计算文档频率
  df_dict = {}
  for word in vocab:
    for doc in docs:
      if word in doc:
        if word in df_dict.keys():
          df_dict[word] += 1
        else:
          df_dict[word] = 1

  # 计算 TF-IDF 值
  tfidf_dict = {}
  for word in vocab:
    tfidf_dict[word] = (tf_dict[word] / sum(tf_dict.values())) * math.log(len(docs) / df_dict[word])

  return tfidf_dict

# 文档集合和词汇表
docs = ["This_is_book", "This_is_apple"]
vocab = ["This_is_apple", "This_is_book"]

tfidf_custom(docs, vocab)