|
1
|
|
|
"""Backend that returns most similar subjects based on similarity in sparse |
|
2
|
|
|
TF-IDF normalized bag-of-words vector space""" |
|
3
|
|
|
|
|
4
|
|
|
import os.path |
|
5
|
|
|
import joblib |
|
6
|
|
|
import gensim.similarities |
|
7
|
|
|
from gensim.matutils import Sparse2Corpus |
|
8
|
|
|
from sklearn.feature_extraction.text import TfidfVectorizer |
|
9
|
|
|
import annif.util |
|
10
|
|
|
from annif.suggestion import VectorSuggestionResult |
|
11
|
|
|
from annif.exception import NotInitializedException, NotSupportedException |
|
12
|
|
|
from . import backend |
|
13
|
|
|
|
|
14
|
|
|
|
|
15
|
|
|
class TFIDFBackend(backend.AnnifBackend): |
|
16
|
|
|
"""TF-IDF vector space similarity based backend for Annif""" |
|
17
|
|
|
name = "tfidf" |
|
18
|
|
|
needs_subject_index = True |
|
19
|
|
|
|
|
20
|
|
|
# defaults for uninitialized instances |
|
21
|
|
|
_vectorizer = None |
|
22
|
|
|
_index = None |
|
23
|
|
|
|
|
24
|
|
|
VECTORIZER_FILE = 'vectorizer' |
|
25
|
|
|
INDEX_FILE = 'tfidf-index' |
|
26
|
|
|
|
|
27
|
|
|
def _initialize_vectorizer(self): |
|
28
|
|
|
if self._vectorizer is None: |
|
29
|
|
|
path = os.path.join(self.datadir, self.VECTORIZER_FILE) |
|
30
|
|
|
if os.path.exists(path): |
|
31
|
|
|
self.debug('loading vectorizer from {}'.format(path)) |
|
32
|
|
|
self._vectorizer = joblib.load(path) |
|
33
|
|
|
else: |
|
34
|
|
|
raise NotInitializedException( |
|
35
|
|
|
"vectorizer file '{}' not found".format(path), |
|
36
|
|
|
backend_id=self.backend_id) |
|
37
|
|
|
|
|
38
|
|
|
def _initialize_index(self): |
|
39
|
|
|
if self._index is None: |
|
40
|
|
|
path = os.path.join(self.datadir, self.INDEX_FILE) |
|
41
|
|
|
self.debug('loading similarity index from {}'.format(path)) |
|
42
|
|
|
if os.path.exists(path): |
|
43
|
|
|
self._index = gensim.similarities.SparseMatrixSimilarity.load( |
|
44
|
|
|
path) |
|
45
|
|
|
else: |
|
46
|
|
|
raise NotInitializedException( |
|
47
|
|
|
'similarity index {} not found'.format(path), |
|
48
|
|
|
backend_id=self.backend_id) |
|
49
|
|
|
|
|
50
|
|
|
def initialize(self): |
|
51
|
|
|
self._initialize_vectorizer() |
|
52
|
|
|
self._initialize_index() |
|
53
|
|
|
|
|
54
|
|
|
def _create_index(self, veccorpus): |
|
55
|
|
|
self.info('creating similarity index') |
|
56
|
|
|
gscorpus = Sparse2Corpus(veccorpus, documents_columns=False) |
|
57
|
|
|
self._index = gensim.similarities.SparseMatrixSimilarity( |
|
58
|
|
|
gscorpus, num_features=len(self._vectorizer.vocabulary_)) |
|
59
|
|
|
annif.util.atomic_save( |
|
60
|
|
|
self._index, |
|
61
|
|
|
self.datadir, |
|
62
|
|
|
self.INDEX_FILE) |
|
63
|
|
|
|
|
64
|
|
|
def train(self, corpus, project): |
|
65
|
|
|
if corpus.is_empty(): |
|
66
|
|
|
raise NotSupportedException( |
|
67
|
|
|
'Cannot train tfidf project with no documents') |
|
68
|
|
|
self.info('transforming subject corpus') |
|
69
|
|
|
subjects = corpus.subjects |
|
70
|
|
|
self.info('creating vectorizer') |
|
71
|
|
|
self._vectorizer = TfidfVectorizer( |
|
72
|
|
|
tokenizer=project.analyzer.tokenize_words) |
|
73
|
|
|
veccorpus = self._vectorizer.fit_transform( |
|
74
|
|
|
(subj.text for subj in subjects)) |
|
|
|
|
|
|
75
|
|
|
annif.util.atomic_save( |
|
76
|
|
|
self._vectorizer, |
|
77
|
|
|
self.datadir, |
|
78
|
|
|
self.VECTORIZER_FILE, |
|
79
|
|
|
method=joblib.dump) |
|
80
|
|
|
self._create_index(veccorpus) |
|
81
|
|
|
|
|
82
|
|
|
def _suggest(self, text, project, params): |
|
83
|
|
|
self.debug('Suggesting subjects for text "{}..." (len={})'.format( |
|
84
|
|
|
text[:20], len(text))) |
|
85
|
|
|
vectors = self._vectorizer.transform([text]) |
|
86
|
|
|
docsim = self._index[vectors[0]] |
|
87
|
|
|
fullresult = VectorSuggestionResult(docsim, project.subjects) |
|
88
|
|
|
return fullresult.filter(limit=int(self.params['limit'])) |
|
89
|
|
|
|
NatLibFi /
Annif
Loading history...