使用Python识别特定名词短语并判断其实际存在性的实现方案

问题概述 #

需要从句子中识别指定目标名词短语，依据上下文的正负表述词（如正面词found/reported，负面词denied/absent）判断该短语是否真实存在，最终输出True或False。示例场景：

当目标短语为red light时：

• s1 = "he reported red light" → True
• s2 = "he found a red light at the end of the road" → True
• s3 = "he denied to have seen red light" → False
• s4 = "he said that the red light was absent" → False

核心问题拆解 #

需解决三个关键场景：

• 长句的分句处理与短语识别
• 多名词短语场景下精准定位目标短语
• 基于正负表述词的存在性逻辑判断

实现方案 #

1. 定义正负触发词库 #

先梳理覆盖常见表述的触发词集合，可根据业务场景扩展：

• 正面存在词（表示短语真实存在）：found, reported, observed, saw, confirmed, noted
• 负面存在词（表示短语不存在）：denied, absent, refuted, lacked, missing
• 否定修饰词（用于抵消正面词的存在性）：not, never, no

2. 结合TextBlob的短语识别与定位 #

利用TextBlob的词性标注和名词短语提取能力，针对长句和多短语场景做以下处理：

• 对长句进行分句遍历，逐个分析分句内的表述与目标短语的关联
• 支持模糊匹配（如a red light匹配red light）或精确匹配，可按需调整

3. 存在性判断逻辑 #

核心逻辑：找到目标短语后，优先检查分句内的负面触发词；若存在正面触发词，需排除被否定词修饰的情况，最终输出判断结果。

完整代码实现 #

from textblob import TextBlob

# 定义正负触发词库，可按需扩展
POSITIVE_TERMS = {"found", "reported", "observed", "saw", "confirmed", "noted"}
NEGATIVE_TERMS = {"denied", "absent", "refuted", "lacked", "missing"}
NEGATION_WORDS = {"not", "never", "no"}

def check_phrase_existence(sentence: str, target_phrase: str) -> bool:
    blob = TextBlob(sentence)
    exists = False
    target_found = False
    
    # 遍历分句处理长句
    for sent in blob.sentences:
        # 提取分句内的名词短语，匹配目标短语（支持包含式匹配）
        noun_phrases = sent.noun_phrases
        if any(target_phrase.lower() in np.lower() for np in noun_phrases):
            target_found = True
            words = [word.lower() for word, _ in sent.pos_tags]
            
            # 先检查是否存在负面触发词，直接返回False
            if any(term in words for term in NEGATIVE_TERMS):
                return False
            
            # 检查正面触发词，并判断是否被否定修饰
            for idx, word in enumerate(words):
                if word in POSITIVE_TERMS:
                    # 限定否定词搜索范围为正面词前3个词，避免跨分句误判
                    if not any(neg in words[max(0, idx-3):idx] for neg in NEGATION_WORDS):
                        exists = True
    
    # 若找到目标短语但无明确触发词，默认返回False（可按需调整默认逻辑）
    return exists if target_found else False

# 测试示例
s1 = "he reported red light"
s2 = "he found a red light at the end of the road"
s3 = "he denied to have seen red light"
s4 = "he said that the red light was absent"
s5 = "he did not find the red light but reported a green light"  # 多短语场景
s6 = "After walking for 10 minutes, she observed a red light turning green, but noted that the crosswalk signal was absent"  # 长句+多短语

print(check_phrase_existence(s1, "red light"))  # True
print(check_phrase_existence(s2, "red light"))  # True
print(check_phrase_existence(s3, "red light"))  # False
print(check_phrase_existence(s4, "red light"))  # False
print(check_phrase_existence(s5, "red light"))  # False
print(check_phrase_existence(s6, "red light"))  # True
print(check_phrase_existence(s6, "crosswalk signal"))  # False

方案优化建议 #

• 触发词扩展：可根据业务领域添加更多专业表述词，比如医疗场景的detected/not detected
• 复杂句式处理：若需处理被动语态、嵌套从句等复杂结构，可替换为spaCy进行依存句法分析，提升精准度
• 匹配逻辑调整：将包含式匹配改为精确匹配，避免相似短语误判（如red light和red lamp）
• 否定词范围优化：根据句子长度动态调整否定词的搜索范围，减少误判概率

内容的提问来源于stack exchange，提问作者A3006