我会这么做：

ser1 = df["search_words"].str.split("\s*,\s*")
ser2 = df["text_to_search"].str.lower()
​
m = [any(w.lower() in t for w in lst) for lst, t in zip(ser1, ser2)]
​
out = df.loc[m]

输出：

print(out)
                          search_words                           text_to_search
0  currency,CrossCurrencySwap,COPTIONS    text1 text2, "text3": TEXT4; Currency
1                          Exotic Bond     text1, text 2 / text 3 'Exotic Bond'
2                            JPN,Japan               Japan, Asia, JPN, EMEA, US
6                              forward  FRAforwardrateagreementforwardbondtrade

这是你想要的吗？

df = pd.DataFrame(
    {
        "search_words": ['word1, word2, word3', 'word4, word5, word6', 'word7, word8, word9'],
        "sentence" : ['this will not match', 'this will match word4', 'this will match word8 word9']
     }
)

df['matched'] = df.apply(lambda row: [search_word for search_word in row['search_words'].split(', ') if search_word in row['sentence']], axis=1)
df[df["matched"].str.len() > 0]

原始DF如下所示：

search_words                     sentence
0  word1, word2, word3          this will not match
1  word4, word5, word6        this will match word4
2  word7, word8, word9  this will match word8 word9

最终df如下所示：

search_words                     sentence         matched
1  word4, word5, word6        this will match word4         [word4]
2  word7, word8, word9  this will match word8 word9  [word8, word9]

编辑：lambda函数中的列表解析看起来很复杂，所以下面是函数格式中的等价物：

def my_function(row: pd.Series):
    search_words = row['search_words'].split(', ')
    result = []
    for search_word in search_words:
        if search_word in row['sentence']:
            result.append(search_word)
    return result

下面是使用apply()执行此操作的两种方法：

• • 方法#1**：将re.search()与从search_words和Series.str.replace()创建的正则表达式模式一起使用，以正则表达式特殊字符|替换逗号分隔符（可能在左右两侧使用空格）：

df2 = pd.DataFrame({
    'search_pat':df.search_words.str.replace('\s*,\s*', '|', regex=True),
    'text_to_search':df.text_to_search})
import re
mask = df2.apply(lambda row: 
    bool(re.search(row.search_pat, row.text_to_search, flags=re.IGNORECASE)), 
    axis=1).tolist()
res = df[mask]

• • 方法#2**：使用Series.str.contains()而不是re.search()创建遮罩：

mask = df2.apply(lambda row: 
    pd.Series(row.text_to_search).str.contains(row.search_pat, case=False).squeeze(),
    axis=1).tolist()

• • 输出**：

search_words                           text_to_search
0  currency,CrossCurrencySwap,COPTIONS    text1 text2, "text3": TEXT4; Currency
1                          Exotic Bond     text1, text 2 / text 3 'Exotic Bond'
2                            JPN,Japan               Japan, Asia, JPN, EMEA, US
6                              forward  FRAforwardrateagreementforwardbondtrade