def test_extract_rules_from_simple_celery_discourse_tree_hierarchical_only_relations(
self):
self._with_celery_text_discourse_tree()
self.rules_extractor = EDUTreeRulesExtractor(
self.discourse_tree, only_hierarchical_relations=False)
rules = self.rules_extractor.extract()
self.assertEqual(len(rules), 6)
def test_build_arrg_graph_rst_conceptnnet_io(self):
self._setup_link_parse_tree_189()
aspects_per_edu = [
(559, [u'phone']), # test added manually
(560, [u'apple']),
(561, []),
(562, [u'store clerk', u'apple', u'teenager', u'advice']),
(563, []),
]
aspects_graph_builder = Aspect2AspectGraph(aspects_per_edu)
rules_extractor = EDUTreeRulesExtractor()
rules = rules_extractor.extract(self.link_tree, [559, 560, 562], 1)
documents_info = {
189: {
'EDUs': [559, 560, 561, 562, 563],
'accepted_edus': [559, 560, 561, 562, 563],
'aspect_concepts': {
189: {
'conceptnet_io': {
u'apple': [
{
'end': u'apple',
'end-lang': u'en',
'relation': u'IsA',
'start': u'object',
'start-lang': u'en',
'weight': 2.8284271
},
{
'end': u'stuff',
'end-lang': u'en',
'relation': u'Synonym',
'start': u'apple',
'start-lang': u'en',
'weight': 2.8284271
},
]
}
}
},
'sentiment': {
559: 1,
560: -1,
5562: 1
},
# the rest of document info is skipped
}
}
graph, page_rank = aspects_graph_builder.build(rules, documents_info,
True)
self.assertEqual(len(rules), 1)
self.assertGreaterEqual(len(graph.nodes()), 4)
self.assertGreaterEqual(len(graph.edges()), 3)
attrib = nx.get_edge_attributes(graph, 'relation_type')
self.assertEqual(
attrib, {
(u'object', u'apple', 0): u'IsA',
(u'apple', u'stuff', 0): u'Synonym',
(u'apple', u'phone', 0): 'Elaboration'
})
def test_build_exemplary_arrg_graph_sample_tree_189_multiaspects(self):
self._setup_link_parse_tree_189()
aspects_per_edu = [
(559, [u'test', u'test2']), # test added manually
(560, [u'thing', u'test2']),
(561, []),
(562, [u'store clerk', u'apple', u'teenager', u'advice']),
(563, [])
]
aspects_graph_builder = Aspect2AspectGraph(
aspects_per_edu, with_cycles_between_aspects=True)
rules_extractor = EDUTreeRulesExtractor()
rules = rules_extractor.extract(self.link_tree, [559, 560, 562], 1)
graph, page_rank = aspects_graph_builder.build(rules,
conceptnet_io=False)
self.assertEqual(len(rules), 1)
self.assertEqual(len(graph.nodes()), 3)
self.assertEqual(len(graph.edges()), 4)
attrib = nx.get_edge_attributes(graph, 'relation_type')
self.assertEqual(
attrib, {
(u'thing', u'test', 0): 'Elaboration',
(u'test2', u'test2', 0): 'Elaboration',
(u'test2', u'test', 0): 'Elaboration',
(u'thing', u'test2', 0): 'Elaboration'
})
def test_multi_aspects_per_edu(self):
self._multi_aspect_per_edu_tree()
rules_extractor = EDUTreeRulesExtractor(
only_hierarchical_relations=False)
rules = rules_extractor.extract(self.discourse_tree,
[559, 560, 561, 562, 563],
doc_id=1)
expected_rules = {
1: [
EDURelation(edu1=560,
edu2=559,
relation_type='Elaboration',
gerani=0.63),
EDURelation(edu1=561,
edu2=559,
relation_type='Elaboration',
gerani=0.53),
EDURelation(edu1=562,
edu2=559,
relation_type='same-unit',
gerani=0.3),
EDURelation(edu1=563,
edu2=559,
relation_type='same-unit',
gerani=0.2),
EDURelation(edu1=561,
edu2=560,
relation_type='Elaboration',
gerani=0.75),
EDURelation(edu1=562,
edu2=560,
relation_type='same-unit',
gerani=0.4),
EDURelation(edu1=563,
edu2=560,
relation_type='same-unit',
gerani=0.3),
EDURelation(edu1=562,
edu2=561,
relation_type='same-unit',
gerani=0.5),
EDURelation(edu1=563,
edu2=561,
relation_type='same-unit',
gerani=0.4),
EDURelation(edu1=563,
edu2=562,
relation_type='Elaboration',
gerani=0.75)
]
}
self.assertEqual(rules, expected_rules)
def test_tree_parsing_and_get_rules_all(self):
rules_extractor = EDUTreeRulesExtractor(
weight_type=['gerani'], only_hierarchical_relations=False)
rules = rules_extractor.extract(self.discourse_tree,
[513, 514, 515, 516, 517], 1)
expected_rules = {
1: [
EDURelation(edu1=514,
edu2=513,
relation_type='Elaboration',
gerani=0.8),
EDURelation(edu1=515,
edu2=513,
relation_type='same-unit',
gerani=0.42),
EDURelation(edu1=516,
edu2=513,
relation_type='same-unit',
gerani=0.33),
EDURelation(edu1=517,
edu2=513,
relation_type='same-unit',
gerani=0.25),
EDURelation(edu1=515,
edu2=514,
relation_type='same-unit',
gerani=0.5),
EDURelation(edu1=516,
edu2=514,
relation_type='same-unit',
gerani=0.42),
EDURelation(edu1=517,
edu2=514,
relation_type='same-unit',
gerani=0.33),
EDURelation(edu1=516,
edu2=515,
relation_type='Elaboration',
gerani=0.6),
EDURelation(edu1=517,
edu2=515,
relation_type='Elaboration',
gerani=0.52),
EDURelation(edu1=517,
edu2=516,
relation_type='Joint',
gerani=0.7)
]
}
self.assertEqual(rules, expected_rules)
def test_bfs_for_several_aspects_in_one_edu(self):
rules_extractor = EDUTreeRulesExtractor(
weight_type=['gerani'], only_hierarchical_relations=True)
rules = rules_extractor.extract(self.discourse_tree,
[513, 514, 515, 516, 517], 1)
expected_rules = {
1: [
EDURelation(edu1=514,
edu2=513,
relation_type='Elaboration',
gerani=0.8),
EDURelation(edu1=516,
edu2=515,
relation_type='Elaboration',
gerani=0.6),
EDURelation(edu1=517,
edu2=515,
relation_type='Elaboration',
gerani=0.52)
]
}
self.assertEqual(rules, expected_rules)
def extract_rules(discourse_tree: nltk.Tree) -> List:
rules_extractor = EDUTreeRulesExtractor(tree=discourse_tree)
return rules_extractor.extract()
def test_load_serialized_tree(self):
self._with_simple_discourse_tree()
self.rules_extractor = EDUTreeRulesExtractor(
self.discourse_tree, only_hierarchical_relations=True)
rules = self.rules_extractor.extract()
self.assertEqual(len(rules), 5)