Spacy依赖匹配器模式不返回匹配项

Question

I am trying to create, add and get results from a pattern using spacy DependencyMatcher.

I created a pattern for the sentence: "From Monday to Friday"

The full pattern:

pattern = [
    {
        "RIGHT_ID": "node0",
        "RIGHT_ATTRS": {'DEP': 'ROOT', 'POS': 'ADP', 'TAG': 'IN'}
    },
    {
        "LEFT_ID": "node0",
        "REL_OP": ">",
        "RIGHT_ID": "node1",
        "RIGHT_ATTRS": {'DEP': 'pobj', 'POS': 'PROPN', 'TAG': 'NNP'},
    },
    {
        "LEFT_ID": "node1",
        "REL_OP": "$--",
        "RIGHT_ID": "node2",
        "RIGHT_ATTRS": {'DEP': 'prep', 'POS': 'ADP', 'TAG': 'IN'},
    },
       {
        "LEFT_ID": "node2",
        "REL_OP": ">",
        "RIGHT_ID": "node3",
        "RIGHT_ATTRS":{'DEP': 'pobj', 'POS': 'PROPN', 'TAG': 'NNP'},
    },
    
]

The simpler pattern is:

pattern = [
    {
        "RIGHT_ID": "node0",
        "RIGHT_ATTRS": {"POS": "ADP"}
    },
    {
        "LEFT_ID": "node0",
        "REL_OP": ">",
        "RIGHT_ID": "node1",
        "RIGHT_ATTRS": {"POS": "PROPN"},
    },
    {
        "LEFT_ID": "node1",
        "REL_OP": "$--",
        "RIGHT_ID": "node2",
        "RIGHT_ATTRS": {"POS": "ADP"},
    },
       {
        "LEFT_ID": "node2",
        "REL_OP": ">",
        "RIGHT_ID": "node3",
        "RIGHT_ATTRS":{'POS': 'PROPN'},
    },
    
]

My question is, why is this pattern not giving any matches, not on the full or simpler pattern?

import spacy
from spacy.matcher import DependencyMatcher


nlp = spacy.load("en_core_web_sm")
matcher = DependencyMatcher(nlp.vocab)


text="From monday to friday"
doc = nlp(text)
matcher.add("pattern1", [pattern])

matches = matcher(doc)

# Each token_id corresponds to one pattern dict
match_id, token_ids = matches[0]

spacy versions:

spaCy v3.0.6

NAME SPACY VERSION

en_core_web_sm >=3.0.0,<3.1.0 3.0.0 ✔

Answer 1

Your REL_OP for node2 is backwards. It should be $++ .

---

To give a full explanation, this code works for me.

import spacy

from spacy.matcher import DependencyMatcher

nlp = spacy.load("en_core_web_sm")
matcher = DependencyMatcher(nlp.vocab)

text="From Monday to Friday"
doc = nlp(text)

pattern = [
    {
        "RIGHT_ID": "node0",
        "RIGHT_ATTRS": {'POS': 'ADP', 'TAG': 'IN'}
    },
    {
        "LEFT_ID": "node0",
        "REL_OP": ">",
        "RIGHT_ID": "node1",
        "RIGHT_ATTRS": {'POS': 'PROPN'},
    },
    {
        "LEFT_ID": "node1",
        "REL_OP": "$++",
        "RIGHT_ID": "node2",
        "RIGHT_ATTRS": {'POS': 'ADP'},
    },
       {
        "LEFT_ID": "node2",
        "REL_OP": ">",
        "RIGHT_ID": "node3",
        "RIGHT_ATTRS":{'POS': 'PROPN'},
    },
    
]

matcher.add("pattern1", [pattern])

matches = matcher(doc)
print(matches)

print("-----")
# this part is just for reference
for word in doc:
    print(word.pos_, word.tag_, word.dep_, word, sep="\t")

Couple of points about this:

• your second pattern is better, you shouldn't need to specify tag and pos for English (tag determines pos)
• In the v3 small model "monday" and "friday" are not proper nouns unless capitalized (it looks like your displaCy output is from the public demo, which uses v2)