SpaCy DependencyMatcher传入Pandas DataFrame列匹配结果为空求助

代码返回空值、结果不符合预期是3个核心错误导致的：

• 匹配执行位置错误：dep_matches = dep_matcher(doc)写在全局作用域，执行时既未定义doc变量，也没有针对DataFrame每行单独生成的doc对象做匹配
• 返回逻辑错误：return rule3_pairs缩进错误，放在for循环内部，匹配到第一个结果就会终止函数，无法拿到同一行文本内的多个动宾组合
• 匹配规则瑕疵：规则里' treat'前面多了冗余空格，installed是动词过去式，对应的词根lemma应该是install，会导致部分词匹配失败

import pandas as pd
import spacy
from spacy.matcher import DependencyMatcher

nlp = spacy.load("en_core_web_lg")
data = {'new':  ['repaired computer and replaced connector.', 'spliced wire on connector.', 'cycled power and reseated connectors and replaced computer on transmitter.']}
df = pd.DataFrame(data)    

# 初始化匹配器、修正规则错误
dep_matcher = DependencyMatcher(vocab=nlp.vocab)
dep_pattern = [
    {
        "RIGHT_ID": "action",
        "RIGHT_ATTRS": {'LEMMA' : {"IN": ["reseat", "cycle", 'replace' , 'repair', 'reinstall' , 'clean', 'treat', 'splice', 'swap', 'read', 'inspect','install' ]}}
    },
    {
        "LEFT_ID": "action",
        "REL_OP": ">",
        "RIGHT_ID": "component",
        "RIGHT_ATTRS": {"DEP":{"IN": ['dobj']}},     
    }
]
dep_matcher.add('maint_action', patterns=[dep_pattern])

def find_matches(text):
    doc = nlp(text.lower()) # 和单字符串测试逻辑对齐，统一转小写
    dep_matches = dep_matcher(doc) # 针对当前行的doc对象执行匹配
    match_res = []
    for match in dep_matches:
        pattern_id, token_ids = match[0], match[1]
        verb_idx, noun_idx = token_ids[0], token_ids[1]
        match_res.append(f"{doc[verb_idx]} {doc[noun_idx]}")
    # 所有匹配完成后统一返回，对齐预期输出格式
    return f"maint_action    {'  '.join(match_res)}"

df['three_tuples'] = df['new'].apply(find_matches)
print(df[['three_tuples']])

执行后输出完全符合预期：

three_tuples
0                  maint_action    repaired computer  replaced connector
1                                maint_action    spliced wire
2  maint_action    cycled power  reseated connectors  replaced computer

后续如果需要扩展匹配规则，只需要修改dep_pattern里的词根、依存关系配置即可，不需要改动匹配执行逻辑。

内容的提问来源于stack exchange，提问作者Robby T