def test_span_tree(question):
tokens = parsing_utils.create_tokens(question.split(" "))
span_tree = skeleton_parser.span_tree_generation_head(tokens=tokens)
# span_tree = skeleton_parser.span_tree_generation_joint(tokens=tokens)
# span_tree_hybrid_dependency_graph = skeleton_to_dependency.span_tree_to_hybrid_dependency_graph_interface(span_tree=span_tree)
print('#question:\t', question)
print('#span tree:\t', span_tree)
return span_tree
def test_ungrounded_graph(question_normal=None):
''' 1. span tree; 2. node mention annotation; 3. dependency tree; 4. relation extraction '''
tokens = parsing_utils.create_tokens(question_normal.split(" "))
span_tree = skeleton_parser.span_tree_generation_head(tokens=tokens)
print('#span tree:', span_tree)
span_tree_hybrid_dependency_graph = skeleton_to_dependency.span_tree_to_hybrid_dependency_graph_interface(span_tree=span_tree)
span_tree_hybrid_dependency_graph = parsing_utils.update_dependencygraph_indexs(old_dependency_graph=span_tree_hybrid_dependency_graph)
ungrounded_nodes = node_recognition.generate_nodes(question_normal=question_normal, qid=None, tokens=tokens)
super_ungrounded_graph = relation_extraction_nff.generate_ungrounded_graph(
ungrounded_nodes=ungrounded_nodes, span_tree_hybrid_dependency_graph=span_tree_hybrid_dependency_graph)
return super_ungrounded_graph
def run_ungrounded_graph_interface(qid=None,
question_normal=None,
gold_graph_query=None,
gold_answer=None,
gold_sparql_query=None):
''' 1. span tree; 2. node mention annotation; 3. dependency tree; 4. relation extraction '''
# 1. span tree generate span tree
tokens = parsing_utils.create_tokens(question_normal.split(" "))
if parsing_args.parser_mode == 'joint':
span_tree = skeleton_parser.span_tree_generation_joint(tokens=tokens)
elif parsing_args.parser_mode == 'head':
span_tree = skeleton_parser.span_tree_generation_head(tokens=tokens)
else:
span_tree = skeleton_parser.span_tree_generation_only_dep(
tokens=tokens)
print('#span tree:', span_tree)
# 3.generate dependency tree
span_tree_hybrid_dependency_graph = skeleton_to_dependency.span_tree_to_hybrid_dependency_graph_interface(
span_tree=span_tree)
span_tree_hybrid_dependency_graph = parsing_utils.update_dependencygraph_indexs(
old_dependency_graph=span_tree_hybrid_dependency_graph)
# 2. node mention annotation sequence_bert_ner_tag_dict
ungrounded_nodes = generated_gold_nodes_by_questions(
question=question_normal)
# aggregation
is_agg, serialization_list = aggregation_interface.aggregation_interface(
question_normal=question_normal)
if is_agg != 'none':
for i, token in enumerate(tokens):
if serialization_list[i] != 'O' and token.ner_tag is None:
token.ner_tag = serialization_list[i]
aggregation_interface.set_class_aggregation_function(
ungrounded_nodes=ungrounded_nodes,
dependency_graph=span_tree_hybrid_dependency_graph,
surface_tokens=tokens)
# bert's ungrounded graph
# ungrounded_graph = ungrounded_graph_generator_bert.generate_ungrounded_graph(ungrounded_nodes=ungrounded_nodes,
# --------------------------- transformation----------------------------------
# relation -> relation ==> {relation} -{class mention} -> {relation} 两个class之间插入空白节点
# ungrounded_graph = transformations.update_merge_relation_to_relation(ungrounded_graph)
# # merge aggregation node
# ungrounded_graph = transformations.update_ungrounded_graph_count_aggregation(ungrounded_graph)
# ungrounded_graph = transformations.update_ungrounded_graph_superlative_aggregation(ungrounded_graph)
# ungrounded_graph = transformations.update_ungrounded_graph_comparative_aggregation(ungrounded_graph)
# #relation -> class; relation -> entity; 转变为 class -{relation}->entity
# ungrounded_graph = transformations.update_ungrounded_graph_contract_relation_node(ungrounded_graph=ungrounded_graph)
# --------------------------------------------------------------------------------
# nff's ungrounded graph
super_ungrounded_graph = relation_extraction_nff.generate_ungrounded_graph(
ungrounded_nodes=ungrounded_nodes,
span_tree_hybrid_dependency_graph=span_tree_hybrid_dependency_graph)
# --------------------------- transformation----------------------------------
# relation -> relation ==> {relation} -{class mention} -> {relation} 两个class之间插入空白节点
# merge aggregation node
# ungrounded_graph = transformations.update_ungrounded_graph_count_aggregation(ungrounded_graph)
# ungrounded_graph = transformations.update_ungrounded_graph_superlative_aggregation(ungrounded_graph)
# ungrounded_graph = transformations.update_ungrounded_graph_comparative_aggregation(ungrounded_graph)
# -------------------------------transformation cwq------------------------------
ungrounded_graphs_list = []
super_ungrounded_graph.set_blag('super')
ungrounded_graphs_list.append(super_ungrounded_graph)
# class{value:wh-word; question_node:1} -> class 转变为class的question_node:1
# 操作1：折叠合并 疑问词节点连接class的话, 折叠疑问词, 连接的class的question node设为1
merge_question_ungrouned_graph = structure_transfers.update_ungrounded_graph_merge_question_node(
ungrounded_graph=super_ungrounded_graph)
if merge_question_ungrouned_graph is not None:
merge_question_ungrouned_graph.set_blag('merge_qc')
ungrounded_graphs_list.append(merge_question_ungrouned_graph)
# 操作2: entity-class entity-{class}-_blank_node s,p,? s, __,o
# ec_ungrounded_graph = transformations.update_ungrounded_graph_ec(ungrounded_graph=super_ungrounded_graph)
# if ec_ungrounded_graph is not None:
# ec_ungrounded_graph.set_blag('adjust_ec')
# ungrounded_graphs_list.append(ec_ungrounded_graph)
# 操作3: 破圈操作: 包含e-e或e-l或l-l的圈，要把它们破开
if merge_question_ungrouned_graph is not None:
current_ungrounded_graph = merge_question_ungrouned_graph
else:
current_ungrounded_graph = super_ungrounded_graph
del_cycle_ungrounded_graph = structure_transfers.undate_ungrounded_graph_del_cycle(
ungrounded_graph=current_ungrounded_graph)
if del_cycle_ungrounded_graph is not None:
del_cycle_ungrounded_graph.set_blag('del_cycle')
ungrounded_graphs_list.append(del_cycle_ungrounded_graph)
# --------------------------------------------------------------------------------
# abstract_question_word = parser_utils.abstract_question_word_generation(tokens=tokens, sequence_ner_tag_dict=sequence_bert_ner_tag_dict)
abstract_question_word = parsing_utils.abstract_question_word_generation(
tokens=tokens, ungrounded_nodes=ungrounded_nodes)
sequence_bert_ner_tag_dict = parsing_utils.get_nertag_sequence(
ungrounded_nodes=ungrounded_nodes)
for ungrounded_graph in ungrounded_graphs_list:
ungrounded_graph.sequence_ner_tag_dict = str(
sequence_bert_ner_tag_dict)
ungrounded_graph.abstract_question = str(abstract_question_word)
ungrounded_graph.important_words_list = str(
parsing_utils.importantwords_by_unimportant_abstractq(
abstract_question_word))
# --------------------------------------------------------------------------------
# generate question annotation
question_annotation = QuestionAnnotation(
qid=qid,
question=question_normal,
question_normal=question_normal,
span_tree=span_tree,
span_tree_hybrid_dependency_graph=span_tree_hybrid_dependency_graph,
super_ungrounded_graph=super_ungrounded_graph,
sequence_ner_tag_dict=sequence_bert_ner_tag_dict,
gold_graph_query=gold_graph_query,
gold_answer=gold_answer,
gold_sparql_query=gold_sparql_query)
# --------------------------------------------------------------------------------
# generate structure
structure = question_annotation.convert_to_structure()
structure.set_ungrounded_graph_forest(
ungrounded_graph_forest=ungrounded_graphs_list)
return structure