|
1
|
|
|
"""Common functionality for analyzers.""" |
|
2
|
|
|
|
|
3
|
|
|
from __future__ import annotations |
|
4
|
|
|
|
|
5
|
|
|
import abc |
|
6
|
|
|
import functools |
|
7
|
|
|
import unicodedata |
|
8
|
|
|
|
|
9
|
|
|
import annif |
|
10
|
|
|
|
|
11
|
|
|
logger = annif.logger |
|
12
|
|
|
|
|
13
|
|
|
_KEY_TOKEN_MIN_LENGTH = "token_min_length" |
|
14
|
|
|
_NLTK_TOKENIZER_DATA = "punkt_tab" |
|
15
|
|
|
|
|
16
|
|
|
|
|
17
|
|
|
class Analyzer(metaclass=abc.ABCMeta): |
|
18
|
|
|
"""Base class for language-specific analyzers. Either tokenize_words or |
|
19
|
|
|
_normalize_word must be overridden in subclasses. Other methods may be |
|
20
|
|
|
overridden when necessary.""" |
|
21
|
|
|
|
|
22
|
|
|
name = None |
|
23
|
|
|
token_min_length = 3 # default value, can be overridden in instances |
|
24
|
|
|
|
|
25
|
|
|
@staticmethod |
|
26
|
|
|
def is_available() -> bool: |
|
27
|
|
|
"""Return True if the analyzer is available for use, False if not.""" |
|
28
|
|
|
return True # can be overridden in implementations if necessary |
|
29
|
|
|
|
|
30
|
|
|
def __init__(self, **kwargs) -> None: |
|
31
|
|
|
if _KEY_TOKEN_MIN_LENGTH in kwargs: |
|
32
|
|
|
self.token_min_length = int(kwargs[_KEY_TOKEN_MIN_LENGTH]) |
|
33
|
|
|
|
|
34
|
|
|
import nltk.data |
|
35
|
|
|
|
|
36
|
|
|
try: |
|
37
|
|
|
nltk.data.find("tokenizers/" + _NLTK_TOKENIZER_DATA) |
|
38
|
|
|
except LookupError as err: |
|
39
|
|
|
logger.debug(str(err)) |
|
40
|
|
|
if _NLTK_TOKENIZER_DATA in str(err): |
|
41
|
|
|
logger.warning( |
|
42
|
|
|
f'NLTK datapackage "{_NLTK_TOKENIZER_DATA}" not found, ' |
|
43
|
|
|
"downloading it now." |
|
44
|
|
|
) |
|
45
|
|
|
nltk.download(_NLTK_TOKENIZER_DATA) |
|
46
|
|
|
else: |
|
47
|
|
|
raise |
|
48
|
|
|
|
|
49
|
|
|
def tokenize_sentences(self, text: str) -> list[str]: |
|
50
|
|
|
"""Tokenize a piece of text (e.g. a document) into sentences.""" |
|
51
|
|
|
import nltk.tokenize |
|
52
|
|
|
|
|
53
|
|
|
return nltk.tokenize.sent_tokenize(text) |
|
54
|
|
|
|
|
55
|
|
|
@functools.lru_cache(maxsize=50000) |
|
56
|
|
|
def is_valid_token(self, word: str) -> bool: |
|
57
|
|
|
"""Return True if the word is an acceptable token.""" |
|
58
|
|
|
if len(word) < self.token_min_length: |
|
59
|
|
|
return False |
|
60
|
|
|
for char in word: |
|
61
|
|
|
category = unicodedata.category(char) |
|
62
|
|
|
if category[0] == "L": # letter |
|
63
|
|
|
return True |
|
64
|
|
|
return False |
|
65
|
|
|
|
|
66
|
|
|
def tokenize_words(self, text: str, filter: bool = True) -> list[str]: |
|
67
|
|
|
"""Tokenize a piece of text (e.g. a sentence) into words. If |
|
68
|
|
|
filter=True (default), only return valid tokens (e.g. not |
|
69
|
|
|
punctuation, numbers or very short words)""" |
|
70
|
|
|
|
|
71
|
|
|
import nltk.tokenize |
|
72
|
|
|
|
|
73
|
|
|
return [ |
|
74
|
|
|
self._normalize_word(word) |
|
75
|
|
|
for word in nltk.tokenize.word_tokenize(text) |
|
76
|
|
|
if (not filter or self.is_valid_token(word)) |
|
77
|
|
|
] |
|
78
|
|
|
|
|
79
|
|
|
def _normalize_word(self, word): |
|
80
|
|
|
"""Normalize (stem or lemmatize) a word form into a normal form.""" |
|
81
|
|
|
pass # pragma: no cover |
|
82
|
|
|
|
NatLibFi /
Annif
