class ComplexMatch:
def __init__(self):
self.data, self.questions = self.load_complex_qa()
self.predictor = Predictor(Path.relation_match_model)
def predict(self, text):
results = []
result = self.predictor.predict_texts(text_a_s=text,
text_b_s=self.questions)
for (qa_type, question), (label, confidence) in zip(self.data, result):
if label == 'No':
confidence = 1 - confidence
results.append((qa_type, question, confidence))
top_5_result = sorted(results, key=lambda x: x[2], reverse=True)[:5]
for qa_type, question, confidence in top_5_result:
logger.debug('{} {} {}'.format(qa_type, question, confidence))
top_qa_type, question, top_confidence = top_5_result[0]
if top_confidence < 0.97:
top_qa_type = ''
return top_qa_type, top_confidence
def load_complex_qa(self):
df = pd.read_excel(os.path.join(Path.data_path, 'complex_qa.xlsx'))
data = []
for _, row in df.iterrows():
data.append((row['qa_type'], row['question']))
questions = [i[1] for i in data]
return data, questions
class RelationClassifier:
"""
关系分类，使用bert进行文本分类
"""
def __init__(self):
self.predictor = Predictor(Path.relation_classifier_model)
def predict(self, text):
label, confidence = self.predictor.predict_text(text_a=text)
logger.debug('{} {} {}'.format(label, confidence, confidence >= 0.6))
if confidence < 0.6:
label = ''
return label, confidence
class RelationMatch:
"""
关系匹配，使用bert文本相似度方法进行
"""
def __init__(self):
self.predicates = self.load_predictes()
self.predictor = Predictor(Path.relation_match_model)
def predict(self, text):
results = []
result = self.predictor.predict_texts(text_a_s=text, text_b_s=self.predicates)
for predicate, (label, confidence) in zip(self.predicates, result):
if label == 'No':
confidence = 1 - confidence
results.append((predicate, confidence))
top_5_result = sorted(results, key=lambda x: x[1], reverse=True)[:5]
for label, confidence in top_5_result:
logger.debug('{} {}'.format(label, confidence))
top_label, top_confidence = top_5_result[0]
if top_confidence < 0.97:
top_label = ''
return top_label, top_confidence
def load_predictes(self):
predicates = []
config_path = os.path.join(Path.dictionary, 'config.txt')
with open(config_path, 'r') as r_f:
for line in r_f:
line = line.rstrip('\n')
if not line:
continue
iri, *words = line.split('\t')
for word in words:
predicates.append(word)
return predicates
import os
import sys
import pandas as pd
from sklearn import metrics
base_path = os.path.join(os.path.dirname(__file__), '..')
sys.path.append(base_path)
from model.text_classification import Predictor
predictor = Predictor(os.path.join(base_path, 'carbot_data/model/kbqapc'))
examples = predictor.processor.get_dev_examples(os.path.join(base_path, 'data/train/kbqa/predicate_classification/'))
data_df = []
for example in examples:
pred, confidence = predictor.predict_text(example.text_a, example.text_b)
text = example.text_a
true = example.label
data_df.append({
'text': text,
'true': true,
'pred': pred,
'是否正确': '是' if true == pred else '否'
})
df = pd.DataFrame(data_df, columns=['text', 'true', 'pred', '是否正确'])
trues = df['true']
preds = df['pred']
labels = sorted(set(trues))
precision, recall, f1, num = metrics.precision_recall_fscore_support(y_true=trues, y_pred=preds, labels=labels,
average=None)
precision_total, recall_total, f1_total, num_total = metrics.precision_recall_fscore_support(y_true=trues, y_pred=preds,
def __init__(self):
self.predicates = self.load_predictes()
self.predictor = Predictor(Path.relation_match_model)
def __init__(self):
self.predictor = Predictor(Path.relation_classifier_model)
def __init__(self):
self.data, self.questions = self.load_complex_qa()
self.predictor = Predictor(Path.relation_match_model)