spacy - PhraseMatcher 用于匹配短语之间的单词

Question

假设我有以下两句话:"Onions are being cut. However, a great big cut to the onions have been observed" ，我希望匹配短语“切洋葱”。这只是一个最小的例子。

我的要求是该算法遍历所有句子并返回该句子是否包含该短语的 bool 值。此外，我只想匹配词形还原版本，短语之间可以有 0 个或多个单词。因此在上面的例子中，我希望它返回 [False, True] .我该怎么做呢？

我的一半尝试如下(我需要帮助的地方标记为 TODO ):

import spacy 
from spacy.matcher import PhraseMatcher 

nlp = spacy.load('en_core_web_sm') 
matcher = PhraseMatcher(nlp.vocab)

corpus = "onions are being cut. However, a great big cut to the onions have been observed"
pattern = "Cutting onions"
doc = nlp(corpus)
# TODO: how do I change the pattern to lemmatize and include any # of words between
matcher.add('pat1', None, pattern) 

results = []
for s in doc.sents:
    # TODO: can I use sentences as a doc?
    matches = matcher(s)
    if len(matches) > 0:
        results.append(True)
    else:
        results.append(False)

Answer 1

我建议使用 spacy.matcher.Matcher并在获得每个匹配后获得匹配的句子。

查看示例演示:

import spacy 
from spacy.matcher import Matcher
​
nlp = spacy.load('en_core_web_sm') 
matcher = Matcher(nlp.vocab)
​
corpus = "onions are being cut. However, a great big cut to the onions have been observed"
doc = nlp(corpus)
pattern = [{'LEMMA': 'cut'},
           {'IS_ALPHA': True, 'OP': '*'},
           {'LEMMA': 'onion'}]
matcher.add('pat', None, pattern) 
​
matches = matcher(doc)
for match_id, start, end in matches:
    string_id = nlp.vocab.strings[match_id]  # Get string representation
    span = doc[start:end]  # The matched span
    print("Match ID: {}\nString ID: {}\nStart: {}\nEnd: {}\nText: {}\nSentence: {}".format(
      match_id, string_id, start, end, span.text, span.sent))

输出:

Match ID: 5387953638794962156
String ID: pat
Start: 10
End: 14
Text: cut to the onions
Sentence: However, a great big cut to the onions have been observed

请注意 pattern = [{'LEMMA': 'cut'},{'IS_ALPHA': True, 'OP': '*'},{'LEMMA': 'onion'}]模式匹配以 cut 开头的字符串引理词( {'LEMMA': 'cut'} )，然后包含任意 0 次或多次出现的任何字母词( {'IS_ALPHA': True, 'OP': '*'} )，然后有一个 onion引理词。