def load_baseline_results(self, dataset='train', force=False):
fname = f'top100_{self.model}_{self.similarity}_{dataset}'
if not force:
results = load_dict_from_json(fname)
if results:
return results
print('Retrieving from index.')
queries = load_dict_from_json(f'{dataset}_set_fixed.json')
if not queries:
print('Cannot find the dataset.')
return None
res = getattr(self, f'baseline_{self.model}_retrieval')(queries)
save_dict_to_json(res, fname)
return res
def get_document_bodies(keyword=None, force=False):
fname = 'document_bodies{}.json'.format('_' + keyword if keyword else '')
if not force:
document_bodies = load_dict_from_json(fname)
if document_bodies:
return document_bodies
print('Creating new entity bodies.')
entities_with_types = get_all_instance_types()
long_abstracts = get_entity_data(
'long_abstracts_en.ttl') if not keyword or keyword == 'long' else {}
short_abstracts = get_entity_data(
'short_abstracts_en.ttl') if not keyword or keyword == 'short' else {}
anchor_texts = get_entity_data(
'anchor_text_en.ttl') if not keyword or keyword == 'anchor' else {}
document_bodies = defaultdict(str)
for entity in entities_with_types:
if (not keyword or keyword == 'short') and entity in short_abstracts:
document_bodies[entity] = short_abstracts[entity]
elif (not keyword or keyword == 'long') and entity in long_abstracts:
document_bodies[entity] = long_abstracts[entity]
if (not keyword or keyword == 'anchor') and entity in anchor_texts:
document_bodies[entity] += anchor_texts[entity]
save_dict_to_json(document_bodies, fname)
print(f'Created {len(document_bodies)} document bodies.')
return document_bodies
def get_all_instance_types(transitive=False, force=False):
fname = f'instance_types{"_all" if transitive else ""}.json'
if not force:
instance_types = load_dict_from_json(fname)
if instance_types:
return instance_types
print("Creating new instance types mapping...")
instance_type_filenames = [
'instance_types_en.ttl', 'instance_types_sdtyped_dbo_en.ttl'
]
if transitive:
instance_type_filenames.append('instance_types_transitive_en.ttl')
instance_types = defaultdict(list)
for filename in instance_type_filenames:
instance_types = get_instance_types(filename, instance_types)
if transitive:
ontology = get_ontology()
for entity in instance_types:
if transitive:
types = []
for t in instance_types[entity]:
types.extend(ontology.get(t, {}).get('path', []))
instance_types[entity] = list(set(types))
else:
instance_types[entity] = list(set(instance_types[entity]))
save_dict_to_json(instance_types, fname)
return instance_types
def remove_non_existing_types(filename, type_hierarchy):
s = load_dict_from_json(filename)
i = 0
for q in s:
if q['category'] == 'resource':
before = len(q['type'])
q['type'] = [t for t in q['type'] if t in type_hierarchy]
if len(q['type']) != before:
i += 1
print(f'Removed extra types from {i} instances.')
save_dict_to_json(s, filename)
def get_type_weights(force=False):
fname = 'type_weight.json'
if not force:
weight_doc = load_dict_from_json(fname)
if weight_doc:
return weight_doc
print('Making new type_weight file.')
types = get_all_instance_types(True)
weight_doc = defaultdict(int)
for _, types in types.items():
for t in types:
weight_doc[t] += 1
save_dict_to_json(weight_doc, fname)
return weight_doc
def get_type_entity(ancestors=False, force=False):
fname = f'type_entity{"_all" if ancestors else ""}.json'
if not force:
documents = load_dict_from_json(fname)
if documents:
return documents
print('Creating new type_entity file.')
entities_with_types = get_all_instance_types(ancestors)
documents = defaultdict(list)
for entity, types in entities_with_types.items():
for t in types:
documents[t].append(entity)
print('Done. Num types: ', len(documents))
save_dict_to_json(documents, fname)
return documents
def get_EC_documents(doc_body='short', force=False):
filename = 'document_EC{}.json'.format('_' + doc_body if doc_body else '')
if not force:
document = load_dict_from_json(filename)
if document:
return document
print('Creating new document.')
bodies = get_document_bodies(doc_body)
# types = get_all_instance_types(transitive=True)
document = defaultdict(dict)
for entity, body in bodies.items():
document[entity]['body'] = body
# document[entity]['type'] = ' '.join(types[entity])
save_dict_to_json(document, filename)
return document
def get_ontology(force=False):
fname = 'ontology.json'
if not force:
ontology = load_dict_from_json(fname)
if ontology:
return ontology
print('Creating new ontology file.')
ontology = defaultdict(dict)
with open(get_data_path('dbpedia_2016-10.nt', True), 'r',
encoding='UTF-8') as f:
for line in f:
if line.startswith('#'):
continue
s, p, o = parse_ttl_line(line)
if (p != 'http://www.w3.org/2000/01/rdf-schema#subClassOf'):
continue
if not ('owl#Thing' in o
or o.startswith('http://dbpedia.org/ontology/')):
continue
subj = resolve_uri(s)
if not subj:
continue
subj = 'dbo:' + subj
obj = resolve_uri(o)
obj = obj if obj == 'owl#Thing' else 'dbo:' + obj
if subj in ontology:
ontology[subj]['parents'].append(obj)
else:
ontology[subj]['parents'] = [obj]
for entity in ontology:
path = list(set(get_path(ontology, entity)))
ontology[entity]['path'] = path
ontology[entity]['num_ancestors'] = len(path)
print(f'There are {len(ontology)} types')
save_dict_to_json(ontology, fname)
return ontology
def get_TC_documents(doc_body='anchor', ancestors=False, force=False):
filename = 'document_TC{}{}.json'.format(
'_' + doc_body if doc_body else '', '_all' if ancestors else '')
if not force:
document = load_dict_from_json(filename)
if document:
return document
print('Creating new document.')
bodies = get_document_bodies(doc_body)
type_entities = get_type_entity(ancestors)
num_types = len(type_entities)
document = defaultdict(dict)
for i, (t, entities) in enumerate(type_entities.items()):
print(
f'Processing {i+1}/{num_types} type with {len(entities)} entities')
document[t]['body'] = ' '.join(
bodies.get(entity, '') for entity in entities)
save_dict_to_json(document, filename)
return document
def remove_empty_queries(filename='train_set_fixed.json'):
s = load_dict_from_json(filename)
num_queries = len(s)
s = [q for q in s if q['question']]
print(f'Removed {num_queries-len(s)} from the dataset')
save_dict_to_json(s, filename)