[ENH] NYTimes - add language to corpus

orangecontrib/text/nyt.py

@@ -4,19 +4,24 @@
 import shelve
 import warnings
 from datetime import date
-from time import sleep
-from urllib import request, parse
+from functools import partial
 from http.client import HTTPException
+from time import sleep
+from urllib import parse, request
 from urllib.error import HTTPError, URLError
 
+import numpy as np
+from dateutil.parser import isoparse
+from Orange.data import (
+    ContinuousVariable,
+    DiscreteVariable,
+    StringVariable,
+    TimeVariable,
+)
+from Orange.misc import environ
 
-from Orange import data
 from orangecontrib.text.corpus import Corpus
-
-try:
-    from Orange.misc import environ
-except ImportError:
-    from Orange.canvas.utils import environ
+from orangecontrib.text.util import create_corpus
 
 SLEEP = 1
 TIMEOUT = 10
@@ -26,39 +31,70 @@
 BASE_URL = 'http://api.nytimes.com/svc/search/v2/articlesearch.json'
 
 
-class NYT:
-    """ Class for fetching records from the NYT API. """
+def keywords(doc, name):
+    kws = doc.get("keywords", [])
+    return ", ".join([kw.get("value") for kw in kws if kw["name"] == name])
 
-    @staticmethod
-    def keywords(doc, name):
-        return ', '.join([kw.get('value')
-                          for kw in doc.get('keywords', [])
-                          if kw['name'] == name])
 
-    attributes = []
+def parse_date(doc):
+    date = doc.get("pub_date")
+    return isoparse(date).timestamp() if date is not None else np.nan
+
+
+class NYT:
+    """ Class for fetching records from the NYT API. """
 
     class_vars = [
-        (data.DiscreteVariable('Section'), lambda doc: doc.get('section_name', None)),
+        (
+            partial(DiscreteVariable, "Section"),
+            lambda doc: doc.get("section_name", None),
+        ),
     ]
 
-    tv = data.TimeVariable('Publication Date')
     metas = [
-        (data.StringVariable('Headline'), lambda doc: doc.get('headline', {}).get('main') or ''),
-        (data.StringVariable('Abstract'), lambda doc: doc.get('abstract') or ''),
-        (data.StringVariable('Snippet'), lambda doc: doc.get('snippet') or ''),
-        (data.StringVariable('Lead Paragraph'), lambda doc: doc.get('lead_paragraph') or ''),
-        (data.StringVariable('Subject Keywords'), lambda doc: NYT.keywords(doc, 'subject')),
-        (data.StringVariable('URL'), lambda doc: doc.get('web_url') or ''),
-        (data.StringVariable('Locations'), lambda doc: NYT.keywords(doc, 'glocations')),
-        (data.StringVariable('Persons'), lambda doc: NYT.keywords(doc, 'persons')),
-        (data.StringVariable('Organizations'), lambda doc: NYT.keywords(doc, 'organizations')),
-        (data.StringVariable('Creative Works'), lambda doc: NYT.keywords(doc, 'creative_works')),
-        (tv, lambda doc: NYT.tv.parse(doc.get('pub_date'))),
-        (data.DiscreteVariable('Article Type'), lambda doc: doc.get('type_of_material', None)),
-        (data.ContinuousVariable('Word Count', number_of_decimals=0), lambda doc: doc.get('word_count', None)),
+        (
+            partial(StringVariable, "Headline"),
+            lambda doc: doc.get("headline", {}).get("main") or "",
+        ),
+        (partial(StringVariable, "Abstract"), lambda doc: doc.get("abstract") or ""),
+        (partial(StringVariable, "Snippet"), lambda doc: doc.get("snippet") or ""),
+        (
+            partial(StringVariable, "Lead Paragraph"),
+            lambda doc: doc.get("lead_paragraph") or "",
+        ),
+        (
+            partial(StringVariable, "Subject Keywords"),
+            partial(keywords, name="subject"),
+        ),
+        (partial(StringVariable, "URL"), lambda doc: doc.get("web_url") or ""),
+        (
+            partial(StringVariable, "Locations"),
+            partial(keywords, name="glocations"),
+        ),
+        (partial(StringVariable, "Persons"), partial(keywords, name="persons")),
+        (
+            partial(StringVariable, "Organizations"),
+            partial(keywords, name="organizations"),
+        ),
+        (
+            partial(StringVariable, "Creative Works"),
+            partial(keywords, name="creative_works"),
+        ),
+        (
+            partial(TimeVariable, "Publication Date", have_time=1, have_date=1),
+            parse_date,
+        ),
+        (
+            partial(DiscreteVariable, "Article Type"),
+            lambda doc: doc.get("type_of_material", None),
+        ),
+        (
+            partial(ContinuousVariable, "Word Count", number_of_decimals=0),
+            lambda doc: doc.get("word_count", None),
+        ),
     ]
 
-    text_features = [metas[0][0], metas[1][0]]  # headline + abstract
+    text_features = ["Headline", "Abstract"]
 
     def __init__(self, api_key):
         """
@@ -129,8 +165,17 @@ def search(self, query, date_from=None, date_to=None, max_docs=None,
         if len(records) > max_docs:
             records = records[:max_docs]
 
-        return Corpus.from_documents(records, 'NY Times', self.attributes,
-                                     self.class_vars, self.metas, title_indices=[-1])
+        corpus = create_corpus(
+            documents=records,
+            attributes=[],
+            class_vars=self.class_vars,
+            metas=self.metas,
+            title_indices=[-1],
+            text_features=self.text_features,
+            name="NY Times",
+        )
+        corpus.attributes["language"] = "en"  # NYT publishes only in English
+        return corpus
 
     def _cache_init(self):
         """ Initialize cache in Orange environment buffer dir. """

orangecontrib/text/tests/test_nyt.py

@@ -7,7 +7,6 @@
 from http.client import HTTPException
 from urllib.error import HTTPError, URLError
 
-from Orange.data import TimeVariable
 from orangecontrib.text import Corpus
 from orangecontrib.text.nyt import NYT, BATCH_SIZE
 
@@ -87,6 +86,7 @@ def test_nyt_key(self):
 
     def test_nyt_query_keywords(self):
         c = self.nyt.search('slovenia', max_docs=10)
+        self.assertEqual(c.language, "en")
         self.assertIsInstance(c, Corpus)
         self.assertEqual(len(c), 10)
 
@@ -96,8 +96,8 @@ def test_nyt_query_date_range(self):
         corpus = self.nyt.search('slovenia', from_date, to_date, max_docs=10)
         self.assertEqual(len(corpus), 10)
 
-        time_index = next(i for i, (var, _) in enumerate(NYT.metas) if isinstance(var, TimeVariable))
-        tv = corpus.domain.metas[time_index]
+        tv = corpus.domain["Publication Date"]
+        time_index = corpus.domain.metas.index(tv)
         for doc in corpus:
             date = tv.repr_val(doc.metas[time_index])
             date = datetime.datetime.strptime(date, '%Y-%m-%d %H:%M:%S').date()
@@ -108,15 +108,20 @@ def test_nyt_query_date_range(self):
     def test_nyt_query_max_records(self):
         c = self.nyt.search('slovenia', max_docs=25)
         self.assertEqual(len(c), 25)
+        self.assertEqual(c.language, "en")
 
     def test_nyt_corpus_domain_generation(self):
         corpus = self.nyt.search('slovenia', max_docs=10)
-        for var, _ in NYT.attributes:
-            self.assertIn(var, corpus.domain.attributes)
-        for var, _ in NYT.class_vars:
-            self.assertIn(var, corpus.domain.class_vars)
-        for var, _ in NYT.metas:
-            self.assertIn(var, corpus.domain.metas)
+        self.assertTupleEqual((), corpus.domain.attributes)
+        self.assertListEqual(
+            [var.args[0] for var, _ in NYT.class_vars],
+            [var.name for var in corpus.domain.class_vars]
+        )
+        self.assertListEqual(
+            [var.args[0] for var, _ in NYT.metas],
+            [var.name for var in corpus.domain.metas]
+        )
+        self.assertEqual(corpus.language, "en")
 
     def test_nyt_result_caching(self):
         self.nyt._fetch_page('slovenia', None, None, 0)     # assure in cache

orangecontrib/text/util.py

@@ -1,6 +1,6 @@
 from functools import wraps
 from math import ceil
-from typing import Union, List, Callable, Any, Tuple
+from typing import Union, List, Callable, Any, Tuple, Optional
 
 import numpy as np
 import scipy.sparse as sp
@@ -102,10 +102,10 @@ def create_corpus(
     title_indices: List[int],
     text_features: List[str],
     name: str,
-    language_attribute: str,
+    language_attribute: Optional[str] = None,
 ):
     """
-    Create a corpus from list of features/documents produced by modelu such as
+    Create a corpus from list of features/documents produced by model such as
     Guardian/NYT
 
     Parameters
@@ -160,7 +160,9 @@ def to_val(attr, val):
     Y = np.array(Y, dtype=np.float64)
     metas = np.array(metas, dtype=object)
 
-    language = infer_language_from_variable(domain[language_attribute])
+    language = None
+    if language_attribute is not None:
+        language = infer_language_from_variable(domain[language_attribute])
     corpus = Corpus.from_numpy(
         domain=domain,
         X=X,

orangecontrib/text/widgets/ownyt.py

@@ -79,7 +79,7 @@ class Outputs:
     date_to = Setting(datetime.now().date())
 
     attributes = [feat.name for feat, _ in NYT.metas if isinstance(feat, StringVariable)]
-    text_includes = Setting([feat.name for feat in NYT.text_features])
+    text_includes = Setting([NYT.text_features])
 
     class Warning(OWWidget.Warning):
         no_text_fields = Msg('Text features are inferred when none are selected.')