使用 NLTK 提取关系

Question

这是follow-up of my question。我正在使用 nltk 来解析人员、组织及其关系。使用this example，我能够创建大量的人和组织；但是，我在 nltk.sem.extract_rel 命令中遇到错误：

AttributeError: 'Tree' object has no attribute 'text'

完整代码如下：

import nltk
import re
#billgatesbio from http://www.reuters.com/finance/stocks/officerProfile?symbol=MSFT.O&officerId=28066
with open('billgatesbio.txt', 'r') as f:
    sample = f.read()

sentences = nltk.sent_tokenize(sample)
tokenized_sentences = [nltk.word_tokenize(sentence) for sentence in sentences]
tagged_sentences = [nltk.pos_tag(sentence) for sentence in tokenized_sentences]
chunked_sentences = nltk.batch_ne_chunk(tagged_sentences)

# tried plain ne_chunk instead of batch_ne_chunk as given in the book
#chunked_sentences = [nltk.ne_chunk(sentence) for sentence in tagged_sentences]

# pattern to find <person> served as <title> in <org>
IN = re.compile(r'.+\s+as\s+')
for doc in chunked_sentences:
    for rel in nltk.sem.extract_rels('ORG', 'PERSON', doc,corpus='ieer', pattern=IN):
        print nltk.sem.show_raw_rtuple(rel)

此示例与given in the book 非常相似，但该示例使用准备好的“已解析文档”，它出现在任何地方，我不知道在哪里可以找到它的对象类型。我也浏览了 git 库。任何帮助表示赞赏。

我的最终目标是提取一些公司的人员、组织、头衔（日期）；然后创建个人和组织的网络地图。

Answer 1

它看起来像是一个“已解析文档”，一个对象需要有一个 headline 成员和一个 text 成员，这两个成员都是令牌列表，其中一些令牌被标记为树。例如，这个（hacky）示例有效：

import nltk
import re

IN = re.compile (r'.*\bin\b(?!\b.+ing)')

class doc():
  pass

doc.headline=['foo']
doc.text=[nltk.Tree('ORGANIZATION', ['WHYY']), 'in', nltk.Tree('LOCATION',['Philadelphia']), '.', 'Ms.', nltk.Tree('PERSON', ['Gross']), ',']

for rel in  nltk.sem.extract_rels('ORG','LOC',doc,corpus='ieer',pattern=IN):
   print nltk.sem.relextract.show_raw_rtuple(rel)

运行时提供输出：

[ORG: 'WHYY'] 'in' [LOC: 'Philadelphia']

显然您实际上不会像这样对其进行编码，但它提供了extract_rels 期望的数据格式的工作示例，您只需要确定如何执行预处理步骤以将您的数据按摩成该格式。

Answer 2

这里是nltk.sem.extract_rels函数的源代码：

def extract_rels(subjclass, objclass, doc, corpus='ace', pattern=None, window=10):
"""
Filter the output of ``semi_rel2reldict`` according to specified NE classes and a filler pattern.

The parameters ``subjclass`` and ``objclass`` can be used to restrict the
Named Entities to particular types (any of 'LOCATION', 'ORGANIZATION',
'PERSON', 'DURATION', 'DATE', 'CARDINAL', 'PERCENT', 'MONEY', 'MEASURE').

:param subjclass: the class of the subject Named Entity.
:type subjclass: str
:param objclass: the class of the object Named Entity.
:type objclass: str
:param doc: input document
:type doc: ieer document or a list of chunk trees
:param corpus: name of the corpus to take as input; possible values are
    'ieer' and 'conll2002'
:type corpus: str
:param pattern: a regular expression for filtering the fillers of
    retrieved triples.
:type pattern: SRE_Pattern
:param window: filters out fillers which exceed this threshold
:type window: int
:return: see ``mk_reldicts``
:rtype: list(defaultdict)
"""
....

因此，如果您将语料库参数作为 ieer 传递，则 nltk.sem.extract_rels 函数期望 doc 参数是 IEERDocument 对象。您应该将语料库作为 ace 传递，或者只是不传递它（默认为 ace）。在这种情况下，它需要一个块树列表（这就是你想要的）。我修改了代码如下：

import nltk
import re
from nltk.sem import extract_rels,rtuple

#billgatesbio from http://www.reuters.com/finance/stocks/officerProfile?symbol=MSFT.O&officerId=28066
with open('billgatesbio.txt', 'r') as f:
    sample = f.read().decode('utf-8')

sentences = nltk.sent_tokenize(sample)
tokenized_sentences = [nltk.word_tokenize(sentence) for sentence in sentences]
tagged_sentences = [nltk.pos_tag(sentence) for sentence in tokenized_sentences]

# here i changed reg ex and below i exchanged subj and obj classes' places
OF = re.compile(r'.*\bof\b.*')

for i, sent in enumerate(tagged_sentences):
    sent = nltk.ne_chunk(sent) # ne_chunk method expects one tagged sentence
    rels = extract_rels('PER', 'ORG', sent, corpus='ace', pattern=OF, window=7) # extract_rels method expects one chunked sentence
    for rel in rels:
        print('{0:<5}{1}'.format(i, rtuple(rel)))

它给出了结果：

[PER: u'Chairman/NNP'] u'and/CC Chief/NNP Executive/NNP Officer/NNP of/IN the/DT' [ORG: u'Company/NNP']

Answer 3

这是 nltk 版本问题。您的代码应该在 nltk 2.x 中工作 但是对于 nltk 3，您应该像这样编写代码

IN = re.compile(r'.*\bin\b(?!\b.+ing)')
for doc in nltk.corpus.ieer.parsed_docs('NYT_19980315'):
    for rel in nltk.sem.relextract.extract_rels('ORG', 'LOC', doc,corpus='ieer', pattern = IN):
         print (nltk.sem.relextract.rtuple(rel))

NLTK Example for Relation Extraction Does not work