class AutoGrocery(object):
"""
"""
def __init__(self, name, train_data):
self._train_data = train_data
self._grocery = Grocery(project_dir + '/models/model_data/' + name)
def train(self):
self._grocery.train(self._train_data)
def save(self):
self._grocery.save()
def load(self):
self._grocery.load()
def predicate(self, src):
if not self._grocery.get_load_status():
try:
self.load()
except ValueError:
self.train()
self.save()
pr = self._grocery.predict(src)
label = pr.predicted_y
return label, pr.dec_values[label]
def tGrocery():
outFile = open('testResult.tmp', 'w')
[trainingSet, benchmark] = pickle.load(open('SampleSeg.pk'))
testingSet = []
correctLabel = []
for i in xrange(len(benchmark)):
print '%d out of %d' % (i, len(benchmark))
testingSet.append(benchmark[i][1])
correctLabel.append(benchmark[i][0])
grocery = Grocery('test')
grocery.train(trainingSet)
grocery.save()
# load
new_grocery = Grocery('test')
new_grocery.load()
Prediction = []
for i in xrange(len(testingSet)):
print '%d out of %d' % (i, len(testingSet))
prediction = new_grocery.predict(testingSet[i])
Prediction.append(prediction)
temp = correctLabel[i] + '<-->' + prediction + ' /x01' + testingSet[i] + '\n'
outFile.write(temp)
correct = 0
for i in xrange(len(Prediction)):
print Prediction[i], correctLabel[i],
if Prediction[i] == correctLabel[i]:
correct += 1
print 'Correct'
else:
print 'False'
print 'Correct Count:', correct
print 'Accuracy: %f' % (1.0 * correct / len(Prediction))
def test(test_path):
new_grocery = Grocery('cv_' + str(fold) +
'_model') #, custom_tokenize=segment)
new_grocery.load()
test_src = []
with open(test_path) as f:
for line in f:
label, text = line.strip().split("|text|")
label = yiji_label[classify_dict[label]]
test_src.append((label, text))
test_result = new_grocery.test(test_src)
#print test_result
#print test_result.accuracy_overall
#accs = test_result.accuracy_labels
recalls = test_result.recall_labels
#print "Recall for each class: ", recalls
predictlabels = test_result.predicted_y
truelabels = test_result.true_y
acc = accuracy_score(truelabels, predictlabels)
macro_precision, macro_recall, macro_fscore, _ = precision_recall_fscore_support(
truelabels, predictlabels, average='macro')
print "Accuracy: ", acc, "Macro-average Precision:", macro_precision, "Macro-average Recall:", macro_recall, "Macro-average Fscore:", macro_fscore
labellist = [
'safe_and_stable', 'industrial_information', 'politics',
'culture_health', 'social_livelihood', 'economic_and_financial'
]
precision, recall, fscore, _ = precision_recall_fscore_support(
truelabels, predictlabels, average=None, labels=labellist)
precisions = dict()
recalls = dict()
for idx, p in enumerate(precision):
precisions[labellist[idx]] = p
for idx, c in enumerate(recall):
recalls[labellist[idx]] = c
def __init__(self, keyword):
print '进行新闻分类'
(db, cursor) = connectdb()
cursor.execute("update task set status=1 where keyword=%s", [keyword])
cursor.execute("select id, title from news where keyword=%s",
[keyword])
news = cursor.fetchall()
new_grocery = Grocery('static/paris')
new_grocery.load()
for item in news:
tag = new_grocery.predict(item['title'])
if tag == '新闻背景':
tag = 1
elif tag == '事实陈述':
tag = 2
elif tag == '事件演化':
tag = 3
elif tag == '各方态度':
tag = 4
elif tag == '直接关联':
tag = 6
elif tag == '暂无关联':
tag = 7
cursor.execute("update news set tag=%s where id=%s",
[tag, item['id']])
closedb(db, cursor)
return
def predict_test(model_path, data):
# 加载模型
try:
model_path = os.path.join(BASE_DIR, 'learn', model_path)
new_grocery = Grocery(model_path.encode('utf-8'))
new_grocery.load()
except Exception as e:
return {'IsErr': True, 'ErrDesc': u'学习模型加载不成功，请检查路径'}
# 整理输入数据
result = list()
sentences = data.split(';')
if sentences[-1] == '':
sentences.pop()
if len(sentences) == 0:
return {'IsErr': True, 'ErrDesc': u'输入的句子结构有错误或没有数据'}
# 分词，再判断
stop_words = read_lines(os.path.join(BASE_DIR, 'learn', 's_w.txt'))
for s in sentences:
tmp_s = ''
words = jieba.cut(s)
for word in words:
if word in stop_words:
continue
else:
tmp_s += word + ' '
result.append({
'tag':
str(new_grocery.predict(tmp_s.strip().encode('utf-8'))),
'sentence':
s,
})
return {'IsErr': False, 'ErrDesc': u'成功', 'data': result}
class GroceryModel(object):
def __init__(self):
self.grocery = Grocery('TextClassify')
def train(self,train_file):
f = open(train_file,'r')
line = f.readline().decode('utf8')
dataset = []
while line:
tmp = line.split('\t')
dataset.append((tmp[0],''.join(tmp[1:])))
line = f.readline().decode('utf8')
f.close()
self.grocery.train(dataset)
self.grocery.save()
def load_model(self):
self.grocery.load()
def test(self,test_src):
self.load_model()
f = open(test_src,'r')
line = f.readline().decode('utf8')
dataset = []
while line:
tmp = line.split('\t')
dataset.append((tmp[0],''.join(tmp[1:])))
line = f.readline().decode('utf8')
f.close()
result = self.grocery.test(dataset)
print result
def predict(self,text):
print self.grocery.predict(text)
def get_data(ids, b_date, end_data, log, stop_word):
b_date = b_date.strftime('%Y-%m-%d')
end_data = end_data.strftime('%Y-%m-%d')
# 选择数据来源
df = load_data(ids, b_date, end_data)
# df = load_data_excel()
# df = pd.read_excel('data_treasure.xls')
df['RateDate'] = pd.to_datetime(df['RateDate'])
# df_group = df['RateDate'].groupby([df.RateDate.values.astype('datetime64[D]')]).size()
res = list()
log.info('Have %d comments need to process' % len(df))
# 分类模型导入
new_grocery = Grocery('sample2')
new_grocery.load()
for record_data in range(0, len(df)):
# 按日期分类摘取内容
# tmp_df = df[df['RateDate'] > df_group.index[record_data]][df['RateDate'] < df_group.index[record_data + 1]]
# 自然语言处理
content_sw, level, tag = nlp_process_with_sw(df.iloc[record_data],
new_grocery, stop_word)
# 记录结果
res.append({
'RateContent': json.dumps(content_sw, ensure_ascii=False),
'RateDate': df.iloc[record_data]['RateDate'],
'TreasureID': df.iloc[record_data]['TreasureID'],
'Level': level,
'Tag': tag,
'Sentence': df.iloc[record_data]['RateContent'],
})
return res
def __init__(self, keyword):
print '进行新闻分类'
(db, cursor) = connectdb()
cursor.execute("update task set status=1 where keyword=%s", [keyword])
cursor.execute("select id, title from news where keyword=%s",[keyword])
news = cursor.fetchall()
new_grocery = Grocery('static/paris')
new_grocery.load()
for item in news:
tag = new_grocery.predict(item['title'])
if tag == '新闻背景':
tag = 1
elif tag == '事实陈述':
tag = 2
elif tag == '事件演化':
tag = 3
elif tag == '各方态度':
tag = 4
elif tag == '直接关联':
tag = 6
elif tag == '暂无关联':
tag = 7
cursor.execute("update news set tag=%s where id=%s", [tag, item['id']])
closedb(db, cursor)
return
def labelmaker(self):
result=[]
grocery = Grocery('11c_20k_20171226')
grocery.load()
label_confidence=sorted(grocery.predict(self.shorttext).dec_values.items(), lambda x, y: cmp(x[1], y[1]), reverse=True)[0]
result.append(label_confidence[0])#置信度最高的分类结果
result.append(label_confidence[1])# 置信度
return result
def test_grocery():
grocery = Grocery('model_redian')
grocery.train('trdata_4.txt')
grocery.save()
new_grocery = Grocery('model_redian')
new_grocery.load()
test_result = new_grocery.test('tedata_4.txt')
print test_result.accuracy_labels
print test_result.recall_labels
test_result.show_result()
def phgrocery(text):
# result_text = []
model_grocery = Grocery('model_redian_5')
model_grocery.load()
result = int(model_grocery.predict(text).predicted_y)
# print result
# if result == 1:
# result_text.append(text)
return result
def test_grocery():
grocery = Grocery('model_redian')
grocery.train('trdata_4.txt')
grocery.save()
new_grocery = Grocery('model_redian')
new_grocery.load()
test_result = new_grocery.test('tedata_4.txt')
print test_result.accuracy_labels
print test_result.recall_labels
test_result.show_result()
def GET(self,name):
#i = web.input(name=None)
#url = "http://"+name
#html = urllib2.urlopen(url).read()
#soup = BeautifulSoup(html)
#title = soup.html.head.title.contents.pop().encode('utf-8')
title = name.encode('utf-8')
new_grocery = Grocery('sample')
new_grocery.load()
return new_grocery.predict(title)
def test_main(self):
grocery = Grocery(self.grocery_name)
grocery.train(self.train_src)
grocery.save()
new_grocery = Grocery('test')
new_grocery.load()
assert grocery.get_load_status()
assert grocery.predict('考生必读：新托福写作考试评分标准') == 'education'
# cleanup
if self.grocery_name and os.path.exists(self.grocery_name):
shutil.rmtree(self.grocery_name)
def predict_phrasing(self, text=u'曾被年轻人嫌弃，如今却媲美Zara'):
'''
:param text:
:param model_name:
:return:
'''
new_grocery = Grocery(self.model_name)
new_grocery.load()
result = new_grocery.predict(text)
return result.dec_values[u'postive']
def test_main(self):
grocery = Grocery(self.grocery_name)
grocery.train(self.train_src)
grocery.save()
new_grocery = Grocery('test')
new_grocery.load()
assert grocery.get_load_status()
assert grocery.predict('考生必读：新托福写作考试评分标准') == 'education'
# cleanup
if self.grocery_name and os.path.exists(self.grocery_name):
shutil.rmtree(self.grocery_name)
def test_main(self):
grocery = Grocery(self.grocery_name)
grocery.train(self.train_src)
grocery.save()
new_grocery = Grocery('test')
new_grocery.load()
assert grocery.get_load_status()
result = grocery.predict('just a testing')
print(result)
result = grocery.predict('考生必读：新托福写作考试评分标准')
print(result)
print("type of result is :",type(result))
assert str(grocery.predict('考生必读：新托福写作考试评分标准')) == 'education'
assert str(grocery.predict('法网')) == 'sports'
# cleanup
if self.grocery_name and os.path.exists(self.grocery_name):
shutil.rmtree(self.grocery_name)
class MyGrocery(object):
def __init__(self, name):
super(MyGrocery, self).__init__()
self.grocery = Grocery(name)
self.loaded = False
self.correct = 1.0
def train(self, src):
lines = []
for line in csv.reader(open(src)):
label, s = line[0],line[1]
text = s.decode('utf8')
lines.append((label, text))
self.grocery.train(lines)
def save_model(self):
self.grocery.save()
def train_and_save(self, src):
self.train(src)
self.save_model()
def load_model(self):
if not self.loaded:
self.grocery.load()
self.loaded = True
def predict(self, text):
self.load_model()
return self.grocery.predict(text)
def test(self, src):
self.load_model()
total, wrong_num = 0.0, 0.0
for line in csv.reader(open(src)):
total += 1
if line[0] != self.predict(line[1]):
wrong_num += 1
print "load test file from " + src
correct = (total - wrong_num ) / total
self.correct = correct
print "total: %d , wrong_num: %d, success percentage: %f" %(total, wrong_num, correct)
result = dict(type="test", total=total, wrong_num=wrong_num, correct=correct)
return json.dumps(result)
def tgrocery_train(train_data,test_data):
'''model预测'''
print("训练语料总数为: " + str(len(train_data)))
test_corpus, test_label = test_split(test_data)
grocery = Grocery('TextGrocery')
print("start training......")
grocery.train(train_data)
grocery.save()
new_grocery = Grocery('TextGrocery')
new_grocery.load()
predict_label = []
for sample in test_corpus:
label = new_grocery.predict(sample)
predict_label.append(str(label))
# print(predict_label)
return test_corpus,test_label,predict_label
def predict_corpus(input_file,output_csv):
import csv
csvfile = file(output_csv, 'wb')
writer = csv.writer(csvfile)
corpus = []
f = xlrd.open_workbook(input_file)
table = f.sheet_by_name('Sheet1')
nrows = table.nrows # 读取行数
for rownum in range(0, nrows):
row = table.row_values(rownum)
row[2].strip()
corpus.append(row[2])
corpus_grocery = Grocery(project_name)
corpus_grocery.load()
output = []
for sentence in corpus:
predict = corpus_grocery.predict(sentence)
output.append((sentence,predict))
writer.writerows(output)
print('Done!')
csvfile.close()
class jdParser(object):
def __init__(self):
self.clf = Grocery("./jdclf")
self.clf.load()
self.LINE_SPLIT = re.compile(u"[；。;\n]")
def get_demand_and_duty(self,jdstr):
linelist = [ line.strip() for line in self.LINE_SPLIT.split(jdstr) if len(line.strip()>4) ]
result = {}
demand = []
duty = []
for line in linelist:
pred = str(self.clf.predict(line))
if pred =="demand":
demand.append(line)
elif pred == "duty":
duty.append(line)
result['demand'] = '\n'.join(demand)
result['duty'] = '\n'.join(duty)
class jdParser(object):
def __init__(self):
self.clf = Grocery("./jdclf")
self.clf.load()
self.LINE_SPLIT = re.compile(u"[；。;\n]")
def get_demand_and_duty(self, jdstr):
linelist = [
line.strip() for line in self.LINE_SPLIT.split(jdstr)
if len(line.strip() > 4)
]
result = {}
demand = []
duty = []
for line in linelist:
pred = str(self.clf.predict(line))
if pred == "demand":
demand.append(line)
elif pred == "duty":
duty.append(line)
result['demand'] = '\n'.join(demand)
result['duty'] = '\n'.join(duty)
class TagPredictor(object):
def _custom_tokenize(self, line, **kwargs):
try:
kwargs["method"]
except:
method = str(self.kwargs["method"])
else:
method = str(kwargs["method"])
if method == "normal":
tokens = self.key_ext.calculateTokens(line,
doc_len_lower_bound=5,
doc_len_upper_bound=500,
method="normal")
elif method == "processed":
tokens = line.split(',')
return tokens
def __init__(self, *args, **kwargs):
self.grocery_name = str(kwargs["grocery_name"])
method = str(kwargs["method"])
train_src = str(kwargs["train_src"])
self.PREFIX = conf.load("predict_label")["prefix"]
self.MODEL_DIR = conf.load("predict_label")["model_dir"]
self.kwargs = kwargs
if method == "normal":
self.key_ext = keyExt()
self.grocery = Grocery(self.grocery_name,
custom_tokenize=self._custom_tokenize)
elif method == "jieba":
self.grocery = Grocery(self.grocery_name)
elif method == "processed":
self.grocery = Grocery(self.grocery_name,
custom_tokenize=self._custom_tokenize)
pass
def trainFromDocs(self, *args, **kwargs):
model = self.grocery.train(self.kwargs["train_src"])
return model
def autoEvaluation(self, *args, **kwargs):
prune_threshold = float(kwargs["threshold"])
excluded_labels = kwargs["excluded_labels"]
excluded_docs = kwargs["excluded_docs"]
train_data = []
with open(self.kwargs["train_src"], 'rb') as f:
for line in f:
try:
line.split('\t', 1)[1]
except:
continue
else:
train_data.append(
(line.split('\t',
1)[0], line.split('\t',
1)[1].split('\n', 1)[0]))
f.close()
print "#items before filtering:", len(train_data)
print "-- Now we filter out the excluded docs --"
train_data = [i for i in train_data if i[1] not in excluded_docs]
print "#items after filtering:", len(train_data)
print "-- Now we filter out the excluded labels --"
train_data = [i for i in train_data if i[0] not in excluded_labels]
print "#items after filtering:", len(train_data)
n = len(train_data) #number of rows in your dataset
indices = range(n)
indices = shuffle(indices)
train_set = map(lambda x: train_data[x], indices[:n * 10 // 10])
test_set = map(lambda x: train_data[x], indices[:n * 10 // 10])
self.grocery.train(train_set)
test_result = self.grocery.test(test_set)
print '-- Accuracy after training --'
print 'Accuracy, A-0:', test_result
low_recall_label = []
for item in test_result.recall_labels.items():
if item[1] < prune_threshold:
low_recall_label.append(item[0])
new_train_set = [
item for item in train_set if item[0] not in low_recall_label
]
new_test_set = [
item for item in train_set if item[0] not in low_recall_label
]
self.grocery.train(new_train_set)
new_test_result = self.grocery.test(new_test_set)
print '-- Accuracy after training, with low-recall labels (less than', str(
prune_threshold * 100), '%) pruned --'
print 'Accuracy, A-1:', new_test_result
return self.grocery, new_test_result
def manualEvaluation(self, *args, **kwargs):
n_docs = int(kwargs["n_docs"])
excluded_labels = kwargs["excluded_labels"]
excluded_docs = kwargs["excluded_docs"]
train_data = []
with open(self.kwargs["train_src"], 'rb') as f:
for line in f:
try:
line.split('\t', 1)[1]
except:
continue
else:
train_data.append(
(line.split('\t',
1)[0], line.split('\t',
1)[1].split('\n', 1)[0]))
f.close()
train_data = [
item for item in train_data if item[0] not in excluded_labels
]
train_data = [i for i in train_data if i[1] not in excluded_docs]
n = len(train_data) #number of rows in your dataset
indices = range(n)
indices = shuffle(indices)
test_set = map(lambda x: train_data[x], indices[0:n_docs])
g = self.loadTrainModel()
test_result = g.test(test_set)
return test_set, test_result
def saveTrainModel(self, *args, **kwargs):
self.grocery.save()
os.rename(
self.PREFIX + self.grocery_name + '_train.svm',
self.PREFIX + self.MODEL_DIR + self.grocery_name + '_train.svm')
return
def loadTrainModel(self, *args, **kwargs):
os.rename(
self.PREFIX + self.MODEL_DIR + self.grocery_name + '_train.svm',
self.PREFIX + self.grocery_name + '_train.svm')
self.grocery.load()
os.rename(
self.PREFIX + self.grocery_name + '_train.svm',
self.PREFIX + self.MODEL_DIR + self.grocery_name + '_train.svm')
return self.grocery
def predict(self, line, **kwargs):
tag = self.grocery.predict(line)
return tag
def test(self, *args, **kwargs):
test_src = str(kwargs["test_src"])
test_result = self.grocery.test(test_src)
print "Total Accuracy", test_result
return test_result
class JdParserTop(object):
def __init__(self):
self.CLEAN_TEXT = re.compile(
u"[^\u4e00-\u9fa5\w\d；:：;,。、\.，。！!@（）\r\n\(\)\-\+ － ]")
self.clf = Grocery(base_dir + "/jdclf")
self.clf.load()
self.SPLIT_LINE = re.compile(u"[\r\n；:：。！？;]|[ \s \xa0\u724b]{4,}")
self.CLEAN_LINE = re.compile(
u"^[\u2022（【\[\s\t\r\n\(\- ]?[\da-z１２３４５７８９]{1,2}[\.,。、，：:）】\]\)\s]|^[！＠＃￥％……＆×（）\(\)｛｝：“｜、－\-，。：:\.]|^[一二三四五六七八九１２３４５６７８９\d]{0,2}[\.、\s:： ]|[，；。、\s \.]$|^[\s \u2022 \uff0d \u25cf]"
)
self.CLEAN_JOBNAME = re.compile(
u"急聘|诚聘|高薪|包[食住宿餐]|.险一金|待遇|^急?招|职位编号\s?[\s\d:：]")
self.PAY = re.compile("(\d{3,}\-)?\d{3,}元")
self.SEX = re.compile(u"性别|男|女")
self.AGE = re.compile(u"\d+周?岁|年龄")
self.JOB_TAG = re.compile(u"全职|实习")
self.DEGREE = re.compile(
u"小学|初中|高中|职技|本科|研究生|硕士|博士|教授|专科|大专|中专|无要求|不限|无限")
self.START_DEMAND = re.compile(
u"(任职资格|岗位要求|工作要求|任职条件|任职要求|职位要求)[：:\s】\n ]?")
self.START_DUTY = re.compile(
u"(工作内容|岗位职责|工作职责|职位描述|工作描述|职位介绍|职位职责|岗位描述)[:：\s 】\n ]")
self.START_BENEFIT = re.compile(u"(福利待遇|待遇|福利)[:：\s\n】]")
self.INC_URL = re.compile(u"(主页|网站|网址|官网).{0,3}[\w\d_/\.:\-]+")
self.DEMAND = re.compile(u"精通|熟悉|熟练|有.+经验")
self.DUTY = re.compile(u"负责|促成|为客户|安排的其.工作")
self.BENEFIT = re.compile(u".险一金|福利|晋身|休假|带薪|补助|补贴")
self.CERT = re.compile(
u"(\S{2,8}证书|CET-\d|普通话|英语|口语|.语|日文|雅思|托福|托业)(至少)?(通过)?[\d一二三四五六七八九]级[及或]?(以上)?|(英语)?CET-\d级?(以上)?|\
医学.{0,3}证|会计.{0,3}证|律师.{0,3}证|有.{1,8}证书")
self.degreedic = set([
line.strip()
for line in codecs.open(base_dir +
'/data/degrees.txt', 'rb', 'utf-8')
])
self.majordic = set([
line.strip()
for line in codecs.open(base_dir +
'/data/majordic.txt', 'rb', 'utf-8')
])
self.skilldic = set([
line.strip()
for line in codecs.open(base_dir +
'/data/skills.txt', 'rb', 'utf-8')
])
self.jobdic = set([
line.strip()
for line in codecs.open(base_dir +
'/data/jobnames.txt', 'rb', 'utf-8')
])
jieba.load_userdict(base_dir + '/data/majordic.txt')
jieba.load_userdict(base_dir + '/data/skills.txt')
jieba.load_userdict(base_dir + '/data/firm.txt')
jieba.load_userdict(base_dir + '/data/degrees.txt')
jieba.load_userdict(base_dir + '/data/benefits.txt')
def clean_line(self, line):
"""
清除一个句子首尾的标点符号
"""
line = self.CLEAN_LINE.sub("", line).strip()
line = re.sub("\s+|^/d+[；’、，／。\.]", "", line)
return line
def clean_cnNum(self, line):
"""
经验年限提取时，中文一二三等转为123
"""
line = unicode(line)
a = [u"一", u"二", u"三", u"四", u"五", u"六", u"七", u"八", u"九", u"十", u"两"]
b = range(1, 11) + [2]
table = dict((ord(aa), bb) for aa, bb in zip(a, b))
return line.translate(table)
def line2vec(self, line):
"""
句子转换为向量
"""
vec = np.zeros(50)
for word in jieba.cut(line):
if word in self.w2v.vocab:
vec += self.w2v[word]
return vec
def clean_jobname(self, jobname):
"""
职位名清洗
"""
if jobname.lower() in self.jobdic:
return jobname.lower()
else:
res = [(lcs_len(jobname, job), job) for job in self.jobdic]
res.sort()
return res[-1][1]
##########################################
# init
model_choose = "svm" # svm, lda, rnn
grocery_name = "./SVM_models/svm_for_news"
corpus_path = "./Corpus/NewsClassCorpus/"
file_path = "./"
file_name = "post.txt"
t_text = delete_stop_words(codecs.open(file_path + file_name, encoding="UTF-8").read())
###########################################
# 调用 SVM 模型分类
if model_choose == "svm":
tic = time.time()
grocery = Grocery(grocery_name)
grocery.load()
t_pre_result = grocery.predict(delete_stop_words(t_text))
toc = time.time()
t_label = t_pre_result.predicted_y
print("Sentiment: ", t_label)
print("How much: ", t_pre_result.dec_values[t_label])
print("Elapsed time of predict is: %s s" % (toc - tic))
elif model_choose == "lda":
pass
elif model_choose == "rnn":
pass
else:
print("")
#!/usr/bin/env python
# -*- coding: utf-8 -*-
import MySQLdb
from tgrocery import Grocery
import sys
reload(sys)
sys.setdefaultencoding('utf8')
grocery = Grocery('sample')
dict_list = list()
conn = MySQLdb.connect(host = 'localhost', db = 'newsdata', user = '******', passwd = 'root', charset = 'utf8', use_unicode = False)
cur = conn.cursor()
cur.execute('select com_new_type_id, com_new_name from tbl_new where com_new_type_id is not null')
for row in cur.fetchall():
dict_list.append(row)
grocery.train(dict_list)
grocery.save()
news_grocery = Grocery('sample')
news_grocery.load()
while True:
result = news_grocery.predict(raw_input('please input title:' ))
print result
def load_first_classifier(model_path):
new_grocery = Grocery(model_path)
new_grocery.load()
return new_grocery
class JdParserTop(object):
def __init__(self):
self.CLEAN_TEXT = re.compile(u"[^\u4e00-\u9fa5\w\d；:：;,。、\.，。！!@（）\r\n\(\)\-\+ － ]")
self.clf = Grocery(base_dir+"/jdclf")
self.clf.load()
self.SPLIT_LINE = re.compile(u"[\r\n；:：。！？;]|[ \s \xa0\u724b]{4,}")
self.CLEAN_LINE = re.compile(u"^[\u2022（【\[\s\t\r\n\(\- ]?[\da-z１２３４５７８９]{1,2}[\.,。、，：:）】\]\)\s]|^[！＠＃￥％……＆×（）\(\)｛｝：“｜、－\-，。：:\.]|^[一二三四五六七八九１２３４５６７８９\d]{0,2}[\.、\s:： ]|[，；。、\s \.]$|^[\s \u2022 \uff0d \u25cf]")
self.CLEAN_JOBNAME = re.compile(u"急聘|诚聘|高薪|包[食住宿餐]|.险一金|待遇|^急?招|职位编号\s?[\s\d:：]")
self.PAY = re.compile("(\d{3,}\-)?\d{3,}元")
self.SEX = re.compile(u"性别|男|女")
self.AGE = re.compile(u"\d+周?岁|年龄")
self.JOB_TAG = re.compile(u"全职|实习")
self.DEGREE = re.compile(u"小学|初中|高中|职技|本科|研究生|硕士|博士|教授|专科|大专|中专|无要求|不限|无限")
self.START_DEMAND = re.compile(u"(任职资格|岗位要求|工作要求|任职条件|任职要求|职位要求)[：:\s】\n ]?")
self.START_DUTY = re.compile(u"(工作内容|岗位职责|工作职责|职位描述|工作描述|职位介绍|职位职责|岗位描述)[:：\s 】\n ]")
self.START_BENEFIT = re.compile(u"(福利待遇|待遇|福利)[:：\s\n】]")
self.INC_URL = re.compile(u"(主页|网站|网址|官网).{0,3}[\w\d_/\.:\-]+")
self.DEMAND = re.compile(u"精通|熟悉|熟练|有.+经验")
self.DUTY = re.compile(u"负责|促成|为客户|安排的其.工作")
self.BENEFIT = re.compile(u".险一金|福利|晋身|休假|带薪|补助|补贴")
self.CERT = re.compile(u"(\S{2,8}证书|CET-\d|普通话|英语|口语|.语|日文|雅思|托福|托业)(至少)?(通过)?[\d一二三四五六七八九]级[及或]?(以上)?|(英语)?CET-\d级?(以上)?|\
医学.{0,3}证|会计.{0,3}证|律师.{0,3}证|有.{1,8}证书")
self.degreedic = set([line.strip() for line in codecs.open(base_dir+'/data/degrees.txt','rb','utf-8')])
self.majordic = set([line.strip() for line in codecs.open(base_dir+'/data/majordic.txt','rb','utf-8')])
self.skilldic = set([line.strip() for line in codecs.open(base_dir+'/data/skills.txt','rb','utf-8')])
self.jobdic = set([line.strip() for line in codecs.open(base_dir+'/data/jobnames.txt','rb','utf-8')])
jieba.load_userdict(base_dir+'/data/majordic.txt')
jieba.load_userdict(base_dir+'/data/skills.txt')
jieba.load_userdict(base_dir+'/data/firm.txt')
jieba.load_userdict(base_dir+'/data/degrees.txt')
jieba.load_userdict(base_dir+'/data/benefits.txt')
def clean_line(self,line):
"""
清除一个句子首尾的标点符号
"""
line = self.CLEAN_LINE.sub("",line).strip()
line = re.sub("\s+|^/d+[；’、，／。\.]","",line)
return line
def clean_cnNum(self,line):
"""
经验年限提取时，中文一二三等转为123
"""
line = unicode(line)
a = [u"一",u"二",u"三",u"四",u"五",u"六",u"七",u"八",u"九",u"十",u"两"]
b = range(1,11)+[2]
table = dict((ord(aa),bb) for aa,bb in zip(a,b))
return line.translate(table)
def line2vec(self,line):
"""
句子转换为向量
"""
vec = np.zeros(50)
for word in jieba.cut(line):
if word in self.w2v.vocab:
vec += self.w2v[word]
return vec
def clean_jobname(self,jobname):
"""
职位名清洗
"""
if jobname.lower() in self.jobdic:
return jobname.lower()
else:
res = [(lcs_len(jobname,job),job) for job in self.jobdic]
res.sort()
return res[-1][1]
class JdParserTop(object):
def __init__(self):
self.result = OrderedDict()
self.result["jdFrom"] = ""
self.result["pubTime"] = ""
inc_keys = [
"incName", "incScale", "incType", "incIndustry", "incLocation",
"incUrl", "incStage", "incAliasName", "investIns",
"incContactInfo", "incCity", "incZipCode", "incContactName",
"incIntro"
]
job_keys = [
"jobType", "jobPosition", "jobCate", "jobSalary", "jobWorkAge",
"jobDiploma", "jobNum", "jobWorkCity", "jobWorkLoc", "jobWelfare",
"age", "jobEndTime", "email", "gender", "jobMajorList", "jobDesc"
]
others_keys = [
"keyWords", "isFullTime", "jdRemedy", "posType", "urgent",
"holidayWelfare", "livingWelfare", "salaryCombine",
"socialWelfare", "trafficWelfare", "jobDepartment", "jobReport",
"jobReportDetail", "jobSubSize", "language", "overSea"
]
jdInc = OrderedDict()
for k in inc_keys:
jdInc[k] = ""
self.result["jdInc"] = jdInc
jdJob = OrderedDict()
for k in job_keys:
jdJob[k] = ""
self.result["jdJob"] = jdJob
others = OrderedDict()
for k in others_keys:
others[k] = ""
self.result["others"] = others
self.CLEAN_TEXT = re.compile(
u"[^\u4e00-\u9fa5\w\d；:：;,。、%\.，/。！!@（）\r\n\(\)\-\+ － `]")
self.clf = Grocery(base_dir + "/jdclf")
self.clf.load()
self.SPLIT_LINE = re.compile(u"[\r\n；:：。！？;]|[ \s \xa0\u724b]{4,}")
self.CLEAN_LINE = re.compile(
u"^[\u2022（【\[\s\t\r\n\(\- ]?[\da-z１２３４５７８９]{1,2}[\.,。、，：:）】\]\)\s]|^[！＠＃￥％……＆×（）\(\)｛｝：“｜、－\-，。：:\.]|^[一二三四五六七八九１２３４５６７８９\d]{0,2}[\.、\s:： ]|[，；。、\s \.]$|^[\s \u2022 \uff0d \u25cf]"
)
self.CLEAN_JOBNAME = re.compile(
u"急聘|诚聘|高薪|包[食住宿餐]|.险一金|待遇|^急?招|职位编号\s?[\s\d:：]")
self.PAY = re.compile("(\d{3,}\-)?\d{3,}元")
self.SEX = re.compile(u"性别|男|女")
self.AGE = re.compile(u"\d+周?岁|年龄")
self.JOB_TAG = re.compile(u"全职|实习")
self.DEGREE = re.compile(
u"小学|初中|高中|职技|本科|研究生|硕士|博士|教授|专科|大专|中专|无要求|不限|无限")
self.MAIL = re.compile(u"\w+@[\w\.]+")
self.START_DEMAND = re.compile(
u"(任职资格|岗位要求|工作要求|任职条件|任职要求|职位要求)[：:\s】\n ]?")
self.START_DUTY = re.compile(
u"(工作内容|岗位职责|工作职责|职位描述|工作描述|职位介绍|职位职责|岗位描述)[:：\s 】\n ]")
self.START_BENEFIT = re.compile(u"(福利待遇|待遇|福利)[:：\s\n】]")
self.INC_URL = re.compile(u"(主页|网站|网址|官网).{0,3}[\w\d_/\.:\-]+")
self.DEMAND = re.compile(u"精通|熟悉|熟练|有.+经验")
self.DUTY = re.compile(u"负责|促成|为客户|安排的其.工作")
self.BENEFIT = re.compile(u".险一金|福利|晋身|休假|带薪|补助|补贴")
self.CERT = re.compile(
u"(\S{2,8}证书|CET-\d|普通话|英语|口语|.语|日文|雅思|托福|托业)(至少)?(通过)?[\d一二三四五六七八九]级[及或]?(以上)?|(英语)?CET-\d级?(以上)?|\
医学.{0,3}证|会计.{0,3}证|律师.{0,3}证|有.{1,8}证书")
self.degreedic = set([
line.strip()
for line in codecs.open(base_dir +
'/data/degrees.txt', 'rb', 'utf-8')
])
self.majordic = set([
line.strip()
for line in codecs.open(base_dir +
'/data/majordic.txt', 'rb', 'utf-8')
])
self.skilldic = set([
line.strip()
for line in codecs.open(base_dir +
'/data/skills.txt', 'rb', 'utf-8')
])
self.jobdic = set([
line.strip()
for line in codecs.open(base_dir +
'/data/jobnames.txt', 'rb', 'utf-8')
])
self.citydic = set([
line.strip()
for line in codecs.open(base_dir +
'/data/citydic.txt', 'rb', 'utf-8')
])
self.province_city = set([
line.strip()
for line in codecs.open(base_dir +
'/data/province_city.txt', 'rb', 'utf-8')
])
self.SALARY = re.compile(u'万')
jieba.load_userdict(base_dir + '/data/majordic.txt')
jieba.load_userdict(base_dir + '/data/skills.txt')
jieba.load_userdict(base_dir + '/data/firm.txt')
jieba.load_userdict(base_dir + '/data/degrees.txt')
jieba.load_userdict(base_dir + '/data/benefits.txt')
jieba.load_userdict(base_dir + '/data/citydic.txt')
jieba.load_userdict(base_dir + '/data/province_city.txt')
def refresh(self):
self.result = OrderedDict()
self.result["jdFrom"] = ""
self.result["pubTime"] = ""
inc_keys = [
"incName", "incScale", "incType", "incIndustry", "incLocation",
"incUrl", "incStage", "incAliasName", "investIns",
"incContactInfo", "incCity", "incZipCode", "incContactName",
"incIntro"
]
job_keys = [
"jobType", "jobPosition", "jobCate", "jobSalary", "jobWorkAge",
"jobDiploma", "jobNum", "jobWorkCity", "jobWorkLoc", "jobWelfare",
"age", "jobEndTime", "email", "gender", "jobMajorList", "jobDesc"
]
others_keys = [
"keyWords", "isFullTime", "jdRemedy", "posType", "urgent",
"holidayWelfare", "livingWelfare", "salaryCombine",
"socialWelfare", "trafficWelfare", "jobDepartment", "jobReport",
"jobReportDetail", "jobSubSize", "language", "overSea"
]
jdInc = OrderedDict()
for k in inc_keys:
jdInc[k] = ""
self.result["jdInc"] = jdInc
jdJob = OrderedDict()
for k in job_keys:
jdJob[k] = ""
self.result["jdJob"] = jdJob
others = OrderedDict()
for k in others_keys:
others[k] = ""
self.result["others"] = others
def clean_line(self, line):
"""
清除一个句子首尾的标点符号
"""
line = self.CLEAN_LINE.sub("", line).strip()
line = re.sub("\s+|^/d+[；’、，／。\.]", "", line)
return line
def clean_cnNum(self, line):
"""
经验年限提取时，中文一二三等转为123
"""
line = unicode(line)
a = [u"一", u"二", u"三", u"四", u"五", u"六", u"七", u"八", u"九", u"十", u"两"]
b = range(1, 11) + [2]
table = dict((ord(aa), bb) for aa, bb in zip(a, b))
return line.translate(table)
def line2vec(self, line):
"""
句子转换为向量
"""
vec = np.zeros(50)
for word in jieba.cut(line):
if word in self.w2v.vocab:
vec += self.w2v[word]
return vec
def clean_jobname(self, jobname):
"""
职位名清洗
"""
print jobname
if jobname.lower() in self.jobdic:
return jobname
else:
res = [(lcs_len(jobname, job), job) for job in self.jobdic]
res.sort()
return res[-1][1]
def desc_extract(self, soup):
line_list = soup.find_all("p")
return '\n'.join([line.get_text() for line in line_list])
#去除img标签,1-7位空格,
removeImg = re.compile('<img.*?>| {1,7}| ')
#删除超链接标签
removeAddr = re.compile('<a.*?>|</a>')
#把换行的标签换为\n
replaceLine = re.compile('<tr>|<div>|</div>|</p>')
#将表格制表<td>替换为\t
replaceTD = re.compile('<td>')
#将换行符或双换行符替换为\n
replaceBR = re.compile('<br><br>|<br>')
#将其余标签剔除
removeExtraTag = re.compile('<.*?>')
#将多行空行删除
removeNoneLine = re.compile('\n+')
def replace(self, x):
x = re.sub(self.removeImg, "", x)
x = re.sub(self.removeAddr, "", x)
x = re.sub(self.replaceLine, "\n", x)
x = re.sub(self.replaceTD, "\t", x)
x = re.sub(self.replaceBR, "\n", x)
x = re.sub(self.removeExtraTag, "", x)
x = re.sub(self.removeNoneLine, "\n", x)
#strip()将前后多余内容删除
return x.strip()
class JdParser(object):
def __init__(self):
self.degreedic = set(line.strip() for line in codecs.open(
'./data/degrees.txt', 'rb', 'utf-8')) # 载入学历词库
self.majordic = set(line.strip() for line in codecs.open(
'./data/majordic.txt', 'rb', 'utf-8')) # 载入专业词库
self.citydic = set(line.strip() for line in codecs.open(
"./data/citydic.txt", 'rb', 'utf-8')) # 载入城市词库
self.firmnames = set(line.strip() for line in codecs.open(
'./data/firm.txt', 'rb', 'utf-8')) # 载入公司缩写名库
self.jobdic = set(line.strip() for line in codecs.open(
'./data/jobposition.txt', 'rb', 'utf-8')) # 载入招聘职位名库
self.skills = set(
line.strip()
for line in codecs.open('./data/skills.txt', 'rb', 'utf-8'))
# self.wordlisttf = pickle.load(open('./data/wordlist.pkl')) # 出现频率最高的2000个单词
# self.w2vdict = json.load(open('./data/word2vec_50.json')) # 2000个词的word2vector
self.clf = Grocery("jdclf") # 句子分类器，分为demand，duty，other
self.clf.load()
self.SEX = re.compile(u"性别不限|性别|男|女")
self.AGE = re.compile(u"\d+周?岁|年龄")
self.DEGREE = re.compile(
u"(全日制)?(初中|高中|中专|大专|专科|大学专科|中职|本科|大学本科|硕士|研究生|博士|博士后)(.?以上)?")
self.MAJOR = re.compile(u"\S+(相关专业|专业优先|及其.专业|[类等]专业[优先]?)")
self.EXP = re.compile(u"工作经验|工作年限|工作经历|项目经[历验]|\d年经[历验]|.{1,2}年相关工作经验")
self.PUB_TIME = re.compile(u"(\d+)(天前发布)")
self.INCNAME = re.compile(
u"\S+(有限公司|酒店|银行|集团|厂|研究中心|研究所|学校|旅行社|中心/s|分?公司|研发中心|技术部|事.部|招聘|商务平台)"
)
self.INCTAG = re.compile(
u"大公司|五百强|全球500强|小公司|成长型公司|创业公司|私有经济|集体经济|集团|外企|已上市|稳定性高|平均年龄\d+岁|妹纸多|学历高|福利待遇好|晋升机会大|民营公司|民营企业|互联网|创业型|国企|央企"
)
self.JOBNAME = re.compile(
u'\S*(研发工程师|工程师|经理|助理|顾问|前台|秘书|主管|研究员|实习生|操作员|专员|教学人员|技术人员|管理员|业务员|公关|程序员|教师|老师|培训生|\
文员|研究员|策划|主任|总监|设计师|分析师|架构师|摄影师|编辑|BD|游戏UI|Android(开发)?|PHP(开发)?|Python(开发)?|.?(急招|急聘|初级|中级|高级|方向).?[\s）】\)])|\
|行政人事|网店设计|客服|会计|电话销售|外贸跟单|web前端|游戏UI|后.开发|产品运营|商业数据分析'
)
self.START_DEMAND = re.compile(
u"(岗位要求|应聘条件|任职要求|岗位资格|任职资格|岗位条件|工作要求|任职条件|人员条件|职位.求|职位条件|职位描述|岗位资格|职位资格|具备条件)[:：\s]\
|如果你.{0,10}[:：\s]|我们希望你.{0,12}[:：\s]|(要求|条件)[:：\s]|你需要?具备什么.+[？\?:：\s]|任职资格[:：\s]"
)
self.DEMAND = re.compile(
u"熟悉|熟练|具有|善于|懂得|掌握|具备|能够|优先|不少于|不超过|至少|团队.作|良好的|工作经验|开发经验|实习经历|能力强|富有|学历|经验|喜欢|较强的.{2,8}能力|相关专业|相关学历|者优先|精通|了解|及以上|技术全面|.强的责任心|[能有]独立|英文流利"
)
self.DUTY = re.compile(
u"跟进|协助|负责|配合|其他工作|领导交办的|对.+提供|审核|参与|为.+提出|日常.+工作|指导|对.+进行|为.+提供|跟进|拓展|运营|用户|客户|协调|拟写|通过|协同|完成|沟通|需求|秘书.{2,5}翻译"
)
self.START_DUTY = re.compile(
u"(岗位职责|岗位描述|职位描述|职责描述|任职描述|职位职责|工作职责|工作职能|职位职能|工作内容|实习内容|职位内容)[:：\s]|做这样的事[:：\s]|职责.{0,5}[:：\s]"
)
self.PAY = re.compile(u"薪酬|待遇|月薪|薪资|年薪|底薪|\d+k|\d+万|\d+元|工资|报酬|薪水|福利")
self.BENEFIT = re.compile(
u"周休|补助|补贴|假日|餐补|提成|交通补助|食宿|加班工资|期权|年假|领导|扁平化|管理|氛围|空间|休假|月假|带薪|全休|晋升|培训|舒适的|旅游|奖励|过节费|五险一金|奖金|\
|弹性工作|氛围|成长空间|实训|培训|高薪|前景|旅游|活动")
self.SPLIT_JD = re.compile(
u"岗位[【（]?[一二三四五六七八九][】）][:：\s]|(^招聘岗位\S+|岗位\d|岗位[一二三四五六])[:：\s]")
self.CLEAR_NUM = re.compile(u"^\d[\.: ：。、]|^[\(（【]?\d[\))】\.]")
self.CLEAR_COLO = re.compile(u"^[\s\.。）（【】，,]|[。；，\.;,]$|^\d[\.]")
self.SKILL = re.compile(
u"精通|了解|熟练|熟悉|掌握|懂得|优先|具备|具有|者优先|擅长|善于|较强的.{2,6}能力|良好的|有.+经验|能力|极强的"
)
jieba.load_userdict('./data/majordic.txt')
jieba.load_userdict('./data/skills.txt')
jieba.load_userdict('./data/firm.txt')
jieba.load_userdict('./data/degrees.txt')
jieba.load_userdict('./data/benefits.txt')
self.jdStr = ""
self.linelist = []
self.lineindex = defaultdict(int)
self.result = OrderedDict()
# 分句，预处理
def preprocess(self, jdstr):
self.result.clear()
jdstr = re.sub(u"[【】◆ \u25cf\u25c6\u2605]", "", jdstr.decode('utf-8'))
self.linelist = [
line.strip() for line in jdstr.split('\n') if len(line) > 1
]
self.jdStr = '\n'.join(self.linelist)
for line in self.linelist:
# print self.clf.predict(line),'\t',line
self.lineindex[re.sub(u"[\s ]+", " ", line)] = 0
def line2vec(self, line):
vec = np.zeros(50)
cnt = 1
for word in jieba.cut(line):
if word in self.w2vdict:
vec += self.w2vdict[word]
cnt += 1
vec = vec / cnt
return vec
# 抽取性别要求
def regular_sex(self):
res = set()
for line in self.linelist:
if self.clf.predict(line) == 'demand' or self.DEMAND.search(line):
findsex = self.SEX.search(line)
if findsex:
getsex = re.search(u"性别不限|男|女",
line.replace(u"男女不限", u"性别不限"))
if getsex:
res.add(getsex.group())
break
if res:
self.result['sex'] = ' / '.join(res)
else:
self.result['sex'] = u'性别不限'
# 抽取年龄要求
def regular_age(self):
res = ''
for line in self.linelist:
if re.search(u'\d{2}后', line): continue
findage = self.AGE.search(line)
if findage:
age = re.findall(u'\d{2}', line)
if len(age) >= 2:
res = '-'.join(age)
elif len(age) == 1:
if re.search(u'以上|不低于', line):
res = age[0] + u'以上'
if re.search(u"不超过|不高于|以下", line):
res = age[0] + '以下'
if re.search(u"左右|大约|大概", line):
res = age[0] + '左右'
break
if len(res) < 2:
res = u'年龄不限'
self.result['age'] = res
return res
# 抽取专业要求
def regular_major(self):
res = []
for line in self.linelist:
findmajor = re.search(u"专业要求[:：\s]", line)
if findmajor:
print 'major demand', line
items = self.clean_line(line[findmajor.span()[1]:]).split()
items = filter(
lambda x: x not in self.degreedic and not re.search(
u"薪酬|经验|元|\d+|月", x), items)
res.append(' / '.join(items))
break
if not res:
for line in self.linelist:
if re.search(u"专业.限|.限专业",
line) and not re.search(u"专业优先", line):
res.append(u"专业不限")
print 'major demand', line
break
else:
findmajor = self.MAJOR.search(line)
if findmajor:
majoritem = re.split(u'[\s,，;； ]', findmajor.group())
for item in majoritem:
if re.search(
u'学历|年龄|岁|学校|公司|性格|具有|具备|能够|经验|有|毕业|性别|男|女',
item):
continue
print 'major item', item
if self.BENEFIT.search(line): continue
print 'major item', item
if re.search(u"专业", item) and len(item) < 3:
continue
res.append(self.clean_line(item))
break
if not res:
for majorword in jieba.cut(line):
if majorword in self.majordic or majorword[:
-2] in self.majordic:
res.append(majorword)
if re.search(u"[等及类]?相关专业",
self.jdStr) and len(res) == 1:
res[0] += u"等相关专业"
if not res:
res.append(u"专业不限")
self.result['major'] = res
# 抽取学历要求
def regular_degree(self):
"""
抽查学历信息，先整找关键字，而后再切词，用词典匹配
"""
degree = [
u'小学', u'初中', u'中专', u'中技', u'高中', u'专科', u'大专', u'本科', u'硕士',
u'博士', u'博士后'
]
res = set()
for line in self.linelist:
finddegree = re.search(u"学历要求[:：\s]", line)
if finddegree:
items = self.clean_line(line[finddegree.span()[1]:]).split()
items = filter(lambda x: not re.search(u"薪酬|经验|元|月|年|\d+", x),
items)
res.add(' / '.join(items))
break
if not res:
for line in self.linelist:
if re.search(u"学历不限|学历要求不限|不限学历", line):
res.add(u"学历不限")
break
else:
finddegree = self.DEGREE.search(line)
if finddegree:
res.add(finddegree.group())
break
# 如果没有匹配到学历的要求信息，就整个文本切词后匹配查找
if len(res) == 0:
for word in jieba.cut(self.jdStr):
if word in self.degreedic:
res.add(word)
res = list(res)
if len(res) == 1 and re.search(u'[及或]?以上', res[0]):
tmp = res[0][:2]
if tmp == u'全日':
tmp = u'本科'
elif tmp == u'研究':
tmp = u'硕士'
if tmp in degree:
idx = degree.index(tmp)
res = degree[idx:]
self.result['degree'] = ' / '.join(res)
# 抽取工作经验年限要求
def regular_exp(self):
cnyear = u'[半一二三四五六七八九十两]年|\d-\d{1,2}年|\d年及?以上|不少于\d年|\d年'
res = set()
jdStr = self.jdStr
findexp = re.search(u'经验不限|(经验)?\d{1,2}年及以上|经验\d-\d{1,2}年', jdStr)
if findexp:
res = findexp.group()
self.result['exp'] = res.replace(u"经验", "")
return res
findexp = self.EXP.search(jdStr)
if findexp:
pos = findexp.span()[1]
jdStr = jdStr[max(0, pos - 25):min(pos + 15, len(jdStr))]
exp = re.search(cnyear, jdStr)
if exp:
res.add(exp.group())
if not res:
exp = re.search(
u"(\d-)?\d{1,2}年(工作|开发|项目)?经[历验]|(不少于)?([半\d]年)及?(以上)?经[历验]|经[历验]\s?(\d-)?\d{1,2}年",
' '.join(self.regular_jobtag()))
if exp:
res.add(exp.group())
else:
exp = re.search(cnyear, ' '.join(self.regular_jobtag()))
if exp:
res.add(exp.group())
self.result["exp"] = "-".join(res)
self.result["exp"] = self.result['exp'].replace(u'经验',
"").replace(u"经历", "")
return res
def regular_jobtag(self):
"""
有关职位标签信息
"""
res = []
job_tag = re.search(u"应届生|全职|兼职|实习生|应届毕业生|社招|急招|急聘", self.jdStr)
if job_tag:
res.append(job_tag.group())
job_tag = re.search(u"招聘人数[:：]?|招聘[:：\s]|人数[：:\s]", self.jdStr)
if job_tag:
jdstr = self.jdStr[job_tag.span()[1]:]
for line in jdstr.split():
if len(line.strip()) < 1: continue
else:
num = re.search(u"(\d+\-)?\d+人?|若干|\d+位", line)
if num:
res.append(u"招聘人数：" + num.group())
break
job_tag = re.search(u"(职能类别|职位标签)[:： ]?", self.jdStr)
if job_tag:
jdstr = self.jdStr[job_tag.span()[1]:]
for line in jdstr.split('\n'):
if len(line.strip()) < 3: continue
else:
res.append("职业标签：" + line.strip())
break
if len(line) > 25: break
# 根据产品部需求专门切割出包含经验的句子等有关职位标注信息,句子进行更精细化切割
linelist = [
line for line in re.split(u"[，。；\s]", self.jdStr)
if 5 < len(line) < 15
]
for line in linelist:
if re.search(u"经验", line) and not re.search(u"月薪|地点|日期", line):
if re.search(u"\d+k|[。？）\)\]]", line): continue
res.append(self.clean_line(line))
break
self.result["job_tag"] = res
return res
# 清除句子前的数字和标点符合
def clean_line(self, line):
line = self.CLEAR_NUM.sub("", line.strip())
line = self.CLEAR_COLO.sub("", line)
return line
# 抽取工作地点
def regular_workplace(self):
res = set()
jdstr = self.jdStr
pos = list(re.finditer(u"(工作地.|上班地.|实习地.|地址|地点)[:：\s]", jdstr))
if pos:
jdstr = jdstr[pos[0].span()[1]:]
for line in jdstr.split():
if len(line.strip()) < 2: continue
if len(line) < 26:
res.add(line.strip().replace(":", "").replace("：", ""))
else:
for city in jieba.cut(line):
if city in self.citydic and city[:-1] not in res:
res.add(city)
break
if not res:
for city in jieba.cut(jdstr):
if city in self.citydic and city[:
-1] not in res and u"国" not in city:
res.add(city)
break
self.result["workplace"] = " / ".join(res)
return res
# 抽取证书获奖情况等其他要求
def regular_cert(self):
res = set()
linelist = [
line for line in re.split(u"[\s ,。；，]", self.jdStr)
if len(line) > 3
]
for line in linelist:
findcert = re.search(
u"(\S+证书|CET-\d|普通话|英语|口语|.语|日文|雅思|托福|托业)(至少)?(通过)?[\d一二三四五六七八九]级[及或]?(以上)?|(英语)?CET-\d级?(以上)?|职业资格|律师证|会计证",
line)
if findcert:
res.add(findcert.group())
else:
findcert = re.search(u"有(.+证)书?", line)
if findcert:
res.add(findcert.group(1))
else:
findcert = re.search(u"有.+资格", line)
if findcert:
res.add(findcert.group())
self.result['cert'] = re.sub(u"[或及以上]", "", ' / '.join(res))
if self.result['cert']:
self.result['cert'] = self.result['cert'].split(' / ')
else:
self.result['cert'] = []
# 利用技能词库抽取相关技能
def regular_skill(self, num=6):
res = []
for line in self.linelist:
if self.DEMAND.search(line) or self.clf.predict(line) == 'demand':
for word in jieba.cut(line):
word = strQ2B(word).lower()
if word in self.skills:
res.append(word)
sorted_words = [w[0] for w in Counter(res).most_common(2 * num)]
for word in jieba.cut(self.result['job_name']):
word = strQ2B(word).lower()
if word in self.skills and word not in sorted_words:
sorted_words.insert(0, word)
after_top3 = sorted_words[3:]
np.random.shuffle(after_top3)
self.result['skill'] = sorted_words[:3] + after_top3[:num - 3]
# 抽取岗位职责
def regular_duty(self):
res = []
jdStr = self.jdStr
pos = list(self.START_DUTY.finditer(jdStr))
if len(pos) > 0:
linelist = [
re.sub("[\s ]+", " ", line)
for line in jdStr[pos[-1].span()[1]:].split("\n")
if len(line) > 2
]
for i in xrange(len(linelist)):
line = linelist[i]
if self.START_DUTY.search(
line) or self.lineindex[line] == 1 or (
re.search(u".年来|谢谢|请在|公司介绍|举报|收藏|岗位职责", line)
and not re.search(u"了解", line)):
continue
if re.search(u"要求[:：\s]?|岗位要求", line) and len(line) < 6: break
if re.match(u"\d{1,2}|\u25cf|[\uff0d（\(\-\+]|[a-z][\.、\s]",
line.strip()) or self.DUTY.search(
line) or self.clf.predict(line) == 'duty':
res.append(line.strip())
elif i < len(linelist) - 1 and self.clf.predict(
linelist[i + 1]) == 'duty':
res.append(line)
else:
break
if not res:
for line in self.linelist:
if re.search(u"粉丝团", line) and len(line) < 12: continue
if self.DUTY.search(line) and self.clf.predict(line) == "duty":
if self.lineindex[line] != 1:
res.append(line)
self.result["duty"] = "\n".join(res)
for line in res:
self.lineindex[line] = 1
return res
# 抽取岗位要求
def regular_demand(self):
res = []
jdStr = self.jdStr
pos = list(self.START_DEMAND.finditer(jdStr))
if len(pos) > 0:
tmppos = pos[-1].span()[0]
if re.search(u"具有|具备", jdStr[tmppos - 5:tmppos + 5]) or re.search(
u"证书|证", jdStr[tmppos:tmppos + 8]):
pos.pop()
if pos:
linelist = [
re.sub("[\s ]+", " ", line)
for line in jdStr[pos[-1].span()[1]:].split("\n")
if len(line) > 2
]
else:
linelist = []
for i in xrange(len(linelist)):
line = linelist[i]
if self.START_DEMAND.search(linelist[i]) or re.search(
u"谢谢|请在|公司介绍|举报|收藏|\d+k?元|加分", line):
continue
if re.match(u"\d{1,2}|\u25cf|[\uff0d（\(\-\+]|[a-z][\.、\s]",
line) or self.DEMAND.search(
line) or self.clf.predict(line) == 'demand':
res.append(line)
elif i < len(linelist) - 1 and self.clf.predict(
linelist[i + 1]) == 'demand':
res.append(line)
else:
break
if not res:
for line in self.linelist:
if self.lineindex[line] == 1 or len(line.split()) > 6:
continue # 如果该句已经被处理过，就不再重复显示
if self.clf.predict(line) == 'demand' or self.DEMAND.search(
line):
res.append(line.strip())
self.result['demand'] = '\n'.join(res)
for line in res:
self.lineindex[line] = 1
return res
# 招聘的职位名
def regular_jobname(self):
res = set()
jdStr = self.jdStr
findpos = re.search(u"(招聘岗位|招聘职位|职位名称|岗位名称|岗位[一二三四五六七八九])[：、:\s ]",
jdStr)
# if not findpos:
# findpos = re.search(u"(职位类别|职位职能)[:：\s ]",jdStr)
if findpos:
pos = findpos.span()[1]
linelist = jdStr[pos:].split("\n")
for line in linelist:
if len(line) < 2: continue
if len(line) >= 2 and len(line) < 20:
if re.search(u"职位描述|查看|地址|工作|分享|举报|下一条|时间|福利|待遇|周末|双休",
line):
continue
res.add(re.sub(u"聘请|高薪诚聘|诚聘|[，。、\d！]+", "", line.strip()))
break
# 如果没有匹配到招聘的具体职位信息，就切词后到职位列表去匹配
if not res:
for line in self.linelist:
if re.search(u"招聘|高薪|诚聘", line): continue
if len(line) < 6 and not re.search(
u'岗位|岗位内容|工作内容|职责|任职|资格',
line) and self.clf.predict(line) == 'job_name':
res.add(line)
break
findPos = self.JOBNAME.search(line)
if findPos and len(findPos.group()) < 20 and not re.match(
u'\d', findPos.group()):
jobname = findPos.group()
res.add(re.sub(u"聘请|高薪诚聘|诚聘|急.|[，。、！]+", "", jobname))
break
# res.add(re.sub(u"\(.+\)|（.+）|【.+】|[，。、\s\d]+|聘请|高薪诚聘|诚聘|急招|","",line.strip()))
if not res:
for line in self.linelist:
for word in jieba.cut(line.lower()):
if word in self.jobdic:
res.add(word)
self.result["job_name"] = " / ".join(res)
return res
if not res:
tag = re.search(u"实习生|兼职", self.jdStr)
if tag:
res.add(tag.group())
self.result["job_name"] = strQ2B(" / ".join(res)).lower()
return res
# 薪酬
def regular_pay(self):
pay = ""
lagoup = re.search(
u"(\d+[kK][-——]\d+[kK])|(\d{3,5}-\d{3,5}元?/[月日天])|(\d{3,5}-\d{3,5}元)|((\d+[-~]\d+)万.[年月])|底薪\d+(-\d+)?元?|\d{3,5}元(左右|以上)?|年薪\d+万?元(左右|以上)?",
self.jdStr) # 针对拉勾网，没有待遇等关键字符
if lagoup:
pay = lagoup.group()
self.result["pay"] = pay
self.result["pay"] = pay.replace(u'k', '000').replace(u'K', '000')
return pay
findpay = self.PAY.search(self.jdStr)
if findpay:
pos = findpay.span()[1]
jdstr = self.jdStr[max(0, pos - 5):min(pos + 10, len(self.jdStr))]
if re.search(u"面议", jdstr):
pay = u"面议"
else:
findpay = re.findall(u"\d{3,7}", jdstr)
pay = "-".join(findpay)
self.result["pay"] = pay.replace(u'k', '000').replace(u'K', '000')
return pay
# 抽取薪资福利
def regular_benefits(self):
res = []
jdStr = self.jdStr
findpos = list(re.finditer(u"薪酬福利[:：\s]|(福利|待遇)\s?[:：]", jdStr))
if not findpos:
findpos = list(
re.finditer(u"(晋升制度|工作环境|职位诱惑|你会获得什么)\s?[？\?:：]", jdStr))
if findpos:
pos = findpos[-1].span()[1]
linelist = jdStr[pos:].split('\n')
for line in linelist:
print 'benefits', line
if len(line.strip()) < 3: continue
if re.match(ur"[（(]?\d+", line) or self.BENEFIT.search(line):
res.append(line.strip())
self.lineindex[line.strip()] = 1
else:
break
if not res:
for line in jdStr.split():
if len(line) > 1 and re.search(
u"带薪|双休|股票期权|五险一金|发展空间|福利|诱惑|休假|薪酬|补助|年假|弹性工作", line):
if re.search(u"福利|待遇|诱惑", line) and len(line.strip()) < 6:
continue
res.append(line.strip())
if len(res) == 1 and re.search(
u"险一金", res[0]) and not re.search(u"[，、]", res[0]):
res[0] = self.clean_line(' '.join(jieba.cut(res[0])))
self.result["benefits"] = "\n".join(res)
return res
# -*- coding: utf-8 -*-
import sys
reload(sys)
sys.path.append('../../')
from config import *
from tgrocery import Grocery
STOP_WORDS_FILE = 'stopwords.txt'
USER_DICT_FILE = 'user_dict.txt'
model_fintext = Grocery('model_fintext')
model_fintext.load()
sys.path.append('../')
from get_es import *
es = Elasticsearch([{'host':ES_HOST,'port':ES_PORT}])
def search(index_name):
es_search_options = set_search_optional()
es_result = get_search_result(es_search_options,index=index_name)
# final_result = get_result_list(es_result)
# return final_result
return es_result
def get_result_list(es_result):
final_result = []
for item in es_result:
final_result.append(item['_source'])
return final_result
def test_sample(path, test_path):
new_grocery = Grocery(path.encode('utf-8'))
new_grocery.load()
test_path = os.path.join(BASE_DIR, 'learn', test_path)
res = new_grocery.test(test_path.encode('utf-8'))
return str(res)
import rospy
from tgrocery import Grocery
from nlu.srv import *
import jieba
import jieba.posseg as pseg
model = '/home/hntea/ros-speech/nlu-model/model'
new_grocery = Grocery(model)
def handle_is_music(req):
# print "Request = [%s]"%(req.topic)
label = str(new_grocery.predict(req.topic))
if label == 'music':
ret = 1
else:
ret = 0
return ret
def is_music_server():
rospy.init_node('nlu_is_music')
s = rospy.Service('is_music', IsMusic, handle_is_music)
print "Servic Is Ready"
rospy.spin()
if __name__ == "__main__":
print "Servic Load Model..."
new_grocery.load() # 加载模型
is_music_server()
# -*- coding: utf-8 -*-
# 分类模型
# Author: Alex
# Created Time: 2016年12月30日 星期五 14时18分58秒
from tgrocery import Grocery
gr = Grocery("test")
gr.load()
lessVal = 0.2
# 统计变量
statDict = {
"total": 0, # 总数
"notEq": 0, # title和content的识别结果不一致
"less": 0, # 小于某个阀值的数量
"title": 0, # 以title为结果的次数
"content": 0 # 以content为结果的次数
}
def getStatDict():
return statDict
def classify(title, content):
"""
分类器
:return cat 预测的分类
# -*- coding: utf-8 -*-
import csv, codecs
from tgrocery import Grocery
import preprocessing as pp
testFileName = '../data/test.txt'
outputFileName = '../output/upload.csv'
# test ##################################
#grocery=Grocery('sample')
grocery = Grocery('version1.0')
grocery.load()
print 'start test'
filetest = codecs.open(testFileName, 'r', 'utf-8')
test_reader = filetest.readlines()
fileOutput = codecs.open(outputFileName, 'w', 'utf-8')
i = 0
for line in test_reader:
content = pp.getcontent(test_reader, i)
i = i + 1
#if(i>10):
#break
if (i % 5000 == 0):
print("%d " % (i)) + '#' * 30
if (content == ''):
print "test.py#" * 3 + line
class JdCRF(object):
def __init__(self):
self.data = []
self.clf = Grocery("jdclf")
self.clf.load()
self.SEX = re.compile(u"性别不限|性别|男|女")
self.AGE = re.compile(u"\d+周?岁|年龄")
self.DEGREE = re.compile(u"(全日制)?(初中|高中|中专|大专|专科|大学专科|中职|本科|大学本科|硕士|研究生|博士|博士后)(.?以上)?")
self.MAJOR = re.compile(u"\S+(相关专业|专业优先|及其.专业|[类等]专业[优先]?)")
self.EXP = re.compile(u"工作经验：|工作经[历验]|工作年限|年.{0,4}经[历验]|经[历验].{1,6}年")
self.PUB_TIME = re.compile(u"(\d+)(天前发布)")
self.INCNAME = re.compile(u"\S+(有限公司|酒店|银行|集团|研究中心|研究所|学校|旅行社|分?公司|研发中心|技术部|事.部|招聘)")
self.NOT_INC = re.compile(u"职位|描述|收藏|推荐|地址|邮箱|主页|介绍|欢迎|加入|要求|简介|险一金|奖金|包吃住|社区|厂房|人员|职责")
self.INCTAG = re.compile(u"大公司|五百强|全球500强|小公司|成长型公司|创业公司|私有经济|集体经济|集团|外企|已上市|稳定性高|平均年龄\d岁|妹纸多|学历高|福利待遇好|晋升机会大|民营公司|民营企业\
|互联网|创业型|国企|央企")
self.JOBNAME = re.compile(u'\S*(研发工程师|工程师|经理|助理|顾问|前台|秘书|主管|研究员|实习生|操作员|专员|教学人员|技术人员|管理员|业务员|公关|程序员|教师|老师|培训生|\
文员|研究员|策划|主任|总监|设计师|分析师|架构师|摄影师|编辑|BD|游戏UI|Android(开发)?|PHP(开发)?|Python(开发)?|.?(急招|急聘|初级|中级|高级|方向).?[\s）】\)])|\
|行政人事|网店设计|客服|会计|电话销售|外贸跟单|web前端|游戏UI|后.开发|产品运营|商业数据分析')
self.START_DEMAND = re.compile(u"(岗位要求|应聘条件|任职要求|岗位资格|任职资格|岗位条件|工作要求|任职条件|人员条件|职位.求|职位条件|职位描述|岗位资格|职位资格|具备条件)[:：\s]\
|如果你.{0,10}[:：\s]|我们希望你.{0,12}[:：\s]|(要求|条件)[:：\s]|你需要?具备什么.+[？\?:：\s]|任职资格[:：\s]")
self.DEMAND = re.compile(u"熟悉|熟练|具有|善于|懂得|掌握|具备|能够|优先|不少于|不超过|至少|团队.作|良好的|工作经验|开发经验|实习经历|能力强|富有|以上学历|经验|喜欢|\
较强的.{2,8}能力|相关专业|相关学历|者优先|精通|了解|及以上|技术全面|.强的责任心|[能有]独立|英文流利")
self.DUTY = re.compile(u"跟进|协助|负责|配合|其他工作|领导交办的|对.+提供|审核|参与|提出|跟踪|报告|为.+提出|日常.+工作|指导|跟进|拓展|运营|用户|客户|协调|拟写|通过|协同\
|完成|沟通|需求|秘书.{2,5}翻译")
self.START_DUTY = re.compile(u"(岗位职责|岗位描述|职位描述|职责描述|任职描述|职位职责|工作职责|工作职能|职位职能|工作内容|实习内容|职位内容)[:：\s]|做这样的事[:：\s]|职责.{0,5}[:：\s]")
self.PAY = re.compile(u"薪酬|待遇|月薪|薪资|年薪|底薪|\d+k|\d+万|\d+元|工资|报酬|薪水|福利")
self.BENEFIT = re.compile(u"周休|补助|补贴|假日|餐补|提成|交通补助|食宿|加班工资|期权|年假|领导|扁平化|管理|氛围|空间|休假|月假|带薪|全休|晋升|培训|舒适的|旅游|奖励|过节费|五险一金|奖金|\
|弹性工作|氛围|成长空间|实训|培训|高薪|前景|旅游|活动|分红")
def gen_data(self,fname='./data/lagou_train.txt'):
fw = codecs.open('./data/jd_train_crf.txt','wb','utf-8')
cnt = 1
for line in codecs.open(fname,'rb','utf-8'):
if line.startswith(u"====="):
fw.write(line)
continue
cnt +=1
if len(line.strip())>1:
pred = self.clf.predict(line)
newline = pred+'\t\t'+line.strip()+'\t\t'+str(len(line))+"\n"
fw.write(newline)
print cnt
print 'done'
def load_data(self,fname="./data/jd_train_crf.txt"):
data = []
tmp = []
for line in codecs.open(fname,'rb','utf-8'):
if line.startswith(u"===="):
data.append(tmp)
tmp = []
continue
else:
tag_data = line.strip().split('\t\t')
if len(tag_data)==3:
tmp.append(tuple(tag_data))
else:
print '\t '.join(tag_data)
n = len(data)/2
print 'train data',n
print 'test data',len(data)-n
return data[n:],data[:n]
def word2features(self,sent,i):
word = sent[i][0]
postag = sent[i][1]
features = [
'bias',
'word.lower=' + word.lower(),
'word[:2]=' +word[:2],
'word.isdigit=%s'%word.isdigit(),
'postag='+postag,
'demand=%s'% '1' if self.DEMAND.search(word) else '0',
'start_demand=%s'% '1' if self.START_DEMAND.search(word) else '0',
'start_duty=%s'% '1' if self.START_DUTY.search(word) else '0',
'duty=%s'% '1' if self.DUTY.search(word) else '0',
'jobname=%s'% '1' if self.JOBNAME.search(word) else '0',
'incname=%s'% '1' if self.INCNAME.search(word) else '0',
'benefit = %s'% '1' if self.BENEFIT.search(word) else '0',
'pred=%s' % self.clf.predict(word)
]
if i>0:
word1 = sent[i-1][0]
postag1 = sent[i-1][1]
features.extend([
'-1:postag='+postag1,
'-1:word.islower='+word1[:3].lower(),
'-1:start_demand=%s'% '1' if self.START_DEMAND.search(word) else '1',
'-1:start_duty=%s'% '1' if self.START_DUTY.search(word) else '0',
'-1:demand=%s'% '1' if self.DEMAND.search(word1) else '0',
'-1:duty=%s'% '1' if self.DUTY.search(word1) else '0',
'-1:jobname=%s'% '1' if self.JOBNAME.search(word1) else '0',
'-1:incname=%s'% '1' if self.INCNAME.search(word1) else '0',
'-1:benefit = %s'% '1' if self.BENEFIT.search(word) else '0',
'-1:pred=%s' % self.clf.predict(word),
])
else:
features.append('BOS')
if i<len(sent)-1:
word1 = sent[i+1][1]
postag1 = sent[i+1][1]
features.extend([
'+1:word.lower=' + word1[:3].lower(),
'+1:word.istitle=%s' % word1.istitle(),
'+1:word.isupper=%s' % word1.isupper(),
'+1:postag=' + postag1,
'+1:postag[:2]=' + postag1[:2],
'+1:start_demand=%s'% '1' if self.START_DEMAND.search(word) else '0',
'+1:start_duty=%s'% '1' if self.START_DUTY.search(word) else '0',
'+1:demand=%s'% '1' if self.DEMAND.search(word1) else '0',
'+1:duty=%s'% '1' if self.DUTY.search(word1) else '0',
'+1:jobname=%s'% '1' if self.JOBNAME.search(word1) else '0',
'+1:incname=%s'% '1' if self.INCNAME.search(word1) else '0',
'+1:benefit = %s'% '1' if self.BENEFIT.search(word) else '0',
'+1:pred=%s' % self.clf.predict(word),
])
else:
features.append('EOS')
return features
def sent2features(self,sent):
return [self.word2features(sent,i) for i in range(len(sent))]
def sent2labels(self,sent):
return [label for (label,token,postag) in sent]
def sent2tokens(self,sent):
return [token for (label,token,postag) in sent]
def train(self,x_train,y_train):
assert len(x_train)==len(y_train),"not the same %d %d"%(len(x_train),len(y_train))
trainer = pycrfsuite.Trainer(verbose=False)
for xseq,yseq in zip(x_train,y_train):
trainer.append(xseq,yseq)
trainer.set_params({
'c1':1.0,
'c2':1e-3,
'max_iterations':50,
'feature.possible_transitions':True
})
trainer.train('jd_skill.crfsuite')
def test(self,sent):
tagger = pycrfsuite.Tagger()
tagger.open('./jd_skill.crfsuite')
print 'tokens ','\n '.join(self.sent2tokens(sent))
print 'Predicted','\t '.join(tagger.tag(self.sent2features(sent)))
print 'Correct ','\t '.join(self.sent2labels(sent))
class XzParserTop(object):
def __init__(self):
self.result = OrderedDict()
inc_keys = [
"jdFrom", "incName", "incAliasName", "incLogo", "incScale",
"incType", "incIndustry", "incIntro", "incCity", "incLocation",
"incZipCode", "incContactName", "incContactInfo", "incUrl"
]
# job_keys = ["pub_time", "jobEndTime", "jobPosition", "jobCate", "jobSalary", "jobWorkAge","jobDiploma", "jobDesc",
# "jobType","jobNum", "jobWorkCity","jobWorkLoc","jobWelfare", "jobMajorList", "age", "gender", "email",
# "jobCVformat", "jobMinimumDays","jobSkill","jobCertificate"]
jdInc = OrderedDict()
for k in inc_keys:
jdInc[k] = ""
self.result["jdInc"] = jdInc
jdJob = OrderedDict()
# for k in job_keys:
# jdJob[k] = ""
self.result["jdJob"] = jdJob
self.CLEAN_TEXT = re.compile(
u"[^\u4e00-\u9fa5\w\d；:：;,。、%\.，/。！!@（）\r\n\(\)\-\+ － `]")
self.clf = Grocery(base_dir + "/jdclf")
self.clf.load()
self.SPLIT_LINE = re.compile(u"[\r\n；:：。！？;]|[ \s \xa0\u724b]{4,}")
self.CLEAN_LINE = re.compile(
u"^[\u2022（【\[\s\t\r\n\(\- ]?[\da-z１２３４５７８９]{1,2}[\.,。、，：:）】\]\)\s]|^[！＠＃￥％……＆×（）\(\)｛｝：“｜、－\-，。：:\.]|^[一二三四五六七八九１２３４５６７８９\d]{0,2}[\.、\s:： ]|[，；。、\s \.]$|^[\s \u2022 \uff0d \u25cf]"
)
self.CLEAN_JOBNAME = re.compile(
u"急聘|诚聘|高薪|包[食住宿餐]|.险一金|待遇|^急?招|职位编号\s?[\s\d:：]")
self.SEX = re.compile(u"性别|男|女")
self.JOB_TAG = re.compile(u"全职|实习")
self.START_DEMAND = re.compile(
u"(任职资格|岗位要求|工作要求|任职条件|任职要求|职位要求)[：:\s】\n ]?")
self.START_DUTY = re.compile(
u"(工作内容|岗位职责|工作职责|职位描述|工作描述|职位介绍|职位职责|岗位描述)[:：\s 】\n ]")
self.START_BENEFIT = re.compile(u"(福利待遇|待遇|福利)[:：\s\n】]")
self.DEMAND = re.compile(u"精通|掌握|熟悉|熟练|有.+经验")
self.DUTY = re.compile(u"负责|促成|为客户|安排的其.工作")
self.BENEFIT = re.compile(u".险一金|福利|晋身|休假|带薪|补助|补贴")
self.CERT = re.compile(
u"(\S{2,8}证书|CET-\d|普通话|英语|口语|.语|日文|雅思|托福|托业)(至少)?(通过)?[\d一二三四五六七八九]级[及或]?(以上)?|(英语)?CET-\d级?(以上)?|\
医学.{0,3}证|会计.{0,3}证|律师.{0,3}证|有.{1,8}证书")
self.degreedic = set([
line.strip()
for line in codecs.open(base_dir +
'/data/degrees.txt', 'rb', 'utf-8')
])
self.majordic = set([
line.strip()
for line in codecs.open(base_dir +
'/data/majordic_new.txt', 'rb', 'utf-8')
])
self.skilldic = set([
line.strip()
for line in codecs.open(base_dir +
'/data/skills.txt', 'rb', 'utf-8')
])
self.jobdic = set([
line.strip()
for line in codecs.open(base_dir +
'/data/jobnames.txt', 'rb', 'utf-8')
])
self.position = set([
line.strip()
for line in codecs.open(base_dir +
'/data/jobposition_new.txt', 'rb', 'utf-8')
])
self.position_prefix = set([
line.strip() for line in codecs.open(
base_dir + '/data/jobposition_prefix.txt', 'rb', 'utf-8')
])
self.position_postfix = set([
line.strip() for line in codecs.open(
base_dir + '/data/jobposition_postfix.txt', 'rb', 'utf-8')
])
self.citydic = set([
line.strip()
for line in codecs.open(base_dir +
'/data/citydic.txt', 'rb', 'utf-8')
])
self.province_city = set([
line.strip()
for line in codecs.open(base_dir +
'/data/province_city.txt', 'rb', 'utf-8')
])
self.city_area = set([
line.strip()
for line in codecs.open(base_dir +
'/data/city_area.txt', 'rb', 'utf-8')
])
jieba.load_userdict(base_dir + '/data/majordic_new.txt')
jieba.load_userdict(base_dir + '/data/skills.txt')
jieba.load_userdict(base_dir + '/data/jobposition.txt')
jieba.load_userdict(base_dir + '/data/firm.txt')
jieba.load_userdict(base_dir + '/data/degrees.txt')
jieba.load_userdict(base_dir + '/data/benefits.txt')
jieba.load_userdict(base_dir + '/data/citydic.txt')
jieba.load_userdict(base_dir + '/data/province_city.txt')
#new
self.INTRO = re.compile(u"公司介绍|公司简介|企业简介|企业介绍|关于我们|单位简介|关于")
self.JOBNAME_LINE = re.compile(
u"岗位：|招聘岗位|实习生招聘|职位名称|招聘职位|实习岗位|岗位方向|定向岗位|岗位$|岗位名称")
self.JOBNAME = re.compile(
u".*?工程师|\S{2}专员$|\S{4,}岗$|工程师$|\S{4,}实习生招聘$|职位\d.*?分析|[^招聘]实习生$|研究员$\
|经理|.*?实习生[((].*?[)）|培训生$]")
self.CONTACTINFO = re.compile(
u'联络方式|联络电话|固定电话|固话|电话|联系电话|QQ|联系方式|传真|Tel')
self.CONTACTNAME = re.compile(u'联络|联系人$|联络人')
self.NUMBER = re.compile(
u'(?<=[^-Q/——_ 0123456789])([-/_ ——0123456789]{7,})')
self.QQ = re.compile(u'QQ\d{6,}|QQ|qq')
self.PUNCTION = re.compile(u'[~\],.;:： ，、。《》【】！#……<>；“”]')
self.INC_URL = re.compile(u"(主页|网站|网址|官网)：(.{0,5}[\w\d_/\.:\-]+)")
self.MAIL = re.compile(u"\w+@[\w\.]+|\w+\.\w+@[\w\.]+|\w.*?at.*?com")
self.FORMART = re.compile(
u'(邮件|简历)名称?以?(主题|格式|标题){0,}为?[~\],.;:：，、。（）《》【】！#……()-<>；“”](.*)|\
("|“){1,}(.*?姓名.*?)["”]|(“|"){1,}(.*?姓名.*?学校.*?)[”"]|("|“){1,}(\S{2,}-\S{2,}-\S{2,}.*?)[”"]|([ 姓名年级学校职位可入职时间-]{5,})'
)
self.JOBLOC = re.compile(u"工作地址|上班地点|公司地址|工作地点|实习地点|总部地址|[^邮件](地址|地点)")
self.MAJOR = re.compile(u"相关专业|等?专业|专业优先|以上学历|优先考虑|专业背景")
self.AGE_LINE = re.compile(u"\d+周?岁|年龄|\d{2}岁")
self.AGE = re.compile(
u"\d{2}?\s?[\- －~到至]?\s?\d{2}周?岁|(至少|不低于|不超过|不大于|大概|大约|不少于|大于)\d+周?岁|\d+周?岁(以上|以下|左右|上下)"
)
self.WEEKTIME = re.compile(
u"(每|一)周(最好)?(至少|最少|保证|起码)?(实习|工作)?[\d一二三四五六七八九十].*?(天|日)(以上)?|实习天数[\d一二三四五六七](天|日)|\
(每|一)周.*?(最好|可到岗)?(至少|最少|保证|起码)?(实习|工作)?[\d一二三四五六七八九十].*?(天|日)(以上)?")
self.JOBLENGTH = re.compile(
u"(实习期)?(至少|保证|起码)?(工作)?[\d一二三四五六七八九十]个.*?月(或)?(以上)|\
周期?.*?(\d个月[-―]{1,2}\d个月)|(实习期)?(至少|保证|起码)(工作)?[\d一二三四五六七八九十]个.*?月(以上)?|至少.{1,5}年(以上)?|(实习).{1,5}年以上"
)
self.XUHAO = re.compile(u"[0123456789一二三四五六七八九]")
self.JOBNUM_LINE = re.compile(u"招?(\d{1,}人)|招聘人数：|岗位人数|人数：")
self.JOBNUM = re.compile(
u"(\(| )?(\d{1,}[人名]?)|(\d{1,}[-_~]?\d{1,}[人名])")
self.PHONE = re.compile(
u'(?<!\d)\d{11,12}(?!\d)|\d{3,4} \d{7,8}|\d{3,4}-\d{7,8}|\d{3,4}—\d{7,8}|\d{3}-\d{3}-\d{4}|(?<!\d)\d{8}(?!\d) '
)
self.ENDTIME = re.compile(
u"(\d.*?[号|日])之?.{0,5}前|截止时间(\d.*日)|截止日期(\d.*日)|(\d.*日)之前")
self.DEGREE_LINE = re.compile(u"(最低)?学历要求|(以上)?学历")
self.DEGREE = re.compile(
u"(小学|初中|高中|职技|本科|研究生|硕士|博士|教授|专科|大专|中专|无要求|不限|Master)(?!优先)|(小学|初中|高中|职技|本科|研究生|硕士|博士|教授|专科|大专|中专|无要求|不限|无限|Master)$"
)
self.SALARY = re.compile(
u"\d{3,}元?\-?\d{3,}元|(本科|研究生|硕士)?\d{2,4}(元|RMB)/?(天|周|月|day|Day)|\dk[--]\dk"
)
#找标题规则
self.FIRST = re.compile(u"一、|二、|三、|四、|五、|六、|七、|八、|九、")
self.THIRD = re.compile(u"[\[【]\S{3,6}[\[】]|^\S{4,5}：$")
#CNN模型
# self.model,self.word_idx_map,self.config = load_model()
#创一个ac自动机用于关键词的匹配。
# builder = AcoraBuilder(list(self.position))
# self.ac = builder.build(builder)
'''
逐行去找信息
self.flag是标志位，主要处理抽取信息跨行的情况：
工作地址：
广东省深圳市XX路
self.extra_info有四个参数：
第一个line是一行文本（不带标签信息）
第二个idx是该行文本所在的行号
第三个add_desc是是否将该行文本加入职位描述信息
第四个clean_major为是否将之前得到的专业与技能信息进行清空
'''
def extra_info(self, line, idx=None, add_desc=True, clean_major=False):
'''
Args:
line (): 输入一行文本
idx (): 该行文本在段落中的行号
add_desc (): 是否将该行信息加入职位描述中，默认True
clean_major (): 是否清空之前的优先专业与技能需求
Returns:
'''
if self.jdType == "regular":
jobName = deepcopy(self.jobName)
elif self.jdType == "special":
jobName = deepcopy(self.jobNameList[idx])
# 加入职位描述
if add_desc:
if self.jdJob[jobName].has_key("jobDesc"):
self.jdJob[jobName]["jobDesc"] += line + u'\n'
else:
self.jdJob[jobName]["jobDesc"] = line + u'\n'
if clean_major:
self.majorList = []
self.skillList = []
# 如果职位名发现人数情况
if self.JOBNUM.search(jobName):
jobNum = re.search(u"\d{1,}人", jobName).group(0)
jobName = jobName.split(jobNum)[0]
self.jdJob[jobName]["jobNum"] = jobNum
if self.flag == "workloc":
self.jdJob[jobName]["jobWorkLoc"] = line
self.flag = None
if self.JOBLOC.search(line):
print 'workloc'
if (len(line) < 20
or len(line) > 100) and not re.search(u"[^城]市|路", line):
pass
elif line.count(u"：") == 1:
workloc = line.split("：")[1]
if len(workloc) > 60:
workloc = re.split(self.PUNCTION, workloc)[0]
self.jdJob[jobName]["jobWorkLoc"] = workloc
if not line.split("：")[1].strip():
self.flag = "workloc"
elif line.count(u"：") > 1:
for tag in filter(None, line.split(" ")):
if self.JOBLOC.search(tag):
if tag.count(u"：") == 1:
self.jdJob[jobName]["jobWorkLoc"] = tag.split(
u"：")[1]
#
elif len(filter(None, self.PUNCTION.split(line))) > 1:
self.jdJob[jobName]["jobWorkLoc"] = filter(
None, self.PUNCTION.split(line))[-1]
else:
self.flag = "workloc"
# 专业、技能采用词库匹配，
if self.DEMAND.search(line):
word_split = jieba.cut(line, cut_all=True)
# 做分词去词库匹配
print "demand"
for word in word_split:
word = word.lower()
if word in self.skilldic:
self.skillList.append(word)
if self.MAJOR.search(line):
word_split = jieba.cut(line)
# 做分词去词库匹配
print "major"
for word in word_split:
word = word.lower()
# print word
word = re.sub(u"相关|学校|专业", u"", word)
if word in self.majordic:
self.majorList.append(word)
if self.FORMART.search(line):
print "format"
if line.count(u"：") == 1 and len(line) < 30:
self.jdJob[jobName]["jobCVformat"] = line.split(u"：")[1]
else:
groups = filter(None, self.FORMART.search(line).groups())
format = groups[np.argmax(map(lambda x: len(x), groups))]
self.jdJob[jobName]["jobCVformat"] = format
if self.ENDTIME.search(line):
print "endtime"
self.jdJob[jobName]["jobEndTime"] = self.ENDTIME.search(
line).group()
if self.MAIL.search(line):
print "email"
if len(self.MAIL.search(line).group()) > 8:
self.jdJob[jobName]["email"] = self.MAIL.search(line).group()
else:
if line.count(u"：") == 1:
self.jdJob[jobName]["email"] = line.split(u"：")[1]
if self.SALARY.search(line):
print "salary"
for item in re.split(u" |；|,|，", line):
if self.SALARY.search(item):
if self.jdJob[jobName].has_key("jobSalary"):
self.jdJob[jobName][
"jobSalary"] += u" " + self.SALARY.search(
item).group()
else:
self.jdJob[jobName]["jobSalary"] = self.SALARY.search(
item).group()
if self.JOBNUM_LINE.search(line):
print "jobnum"
self.jdJob[jobName]["jobNum"] = self.JOBNUM.search(line).group()
if self.WEEKTIME.search(line):
print "weektime"
self.jdJob[jobName]["jobMinimumDays"] = self.WEEKTIME.search(
line).group()
if self.JOBLENGTH.search(line):
print "jobLength"
self.jdJob[jobName]["jobLength"] = self.JOBLENGTH.search(
line).group()
if self.DEGREE.search(line):
print "degree"
line = re.sub(u"士研究生", u"士", line)
print filter(lambda x: len(x) > 1, self.DEGREE.findall(line))
self.jdJob[jobName]["jobDiploma"] = list(
set(filter(None,
self.DEGREE.findall(line)[0])))
if self.AGE_LINE.search(line):
print "age"
findage = self.AGE.search(line)
if findage:
self.jdJob[jobName]["age"] = findage.group()
if len(self.majorList) > 0:
self.jdJob[jobName]["jobMajorList"] = list(set(self.majorList))
if len(self.skillList) > 0:
self.jdJob[jobName]["jobSkill"] = list(set(self.skillList))
# elif self.WEEKTIME.search(line):
# print "worktime"
# if line.count(u"：") == 2:
# worktime = self.CLEAN_TEXT.sub(u"", line.split(u"：")[1])
# self.jdJob[jobName]["jobMinimumDays"] = worktime
# if not worktime:
# self.flag = "worktime"
# else:
#
# if self.flag == "worktime":
# self.jdJob[jobName]["jobMinimumDays"] += self.WORKTIME.search(line).group(0)
# else:
# if self.WEEKTIME.search(line).group(0).find(u"：")<2:
# self.jdJob[jobName]["jobMinimumDays"] = self.WORKTIME.search(line).group(0)
def refresh(self):
self.result = OrderedDict()
inc_keys = [
"jdFrom", "incName", "incAliasName", "incLogo", "incScale",
"incType", "incIndustry", "incIntro", "incCity", "incLocation",
"incZipCode", "incContactName", "incContactInfo", "incUrl"
]
# job_keys = ["pub_time", "jobEndTime", "jobPosition", "jobCate", "jobSalary", "jobWorkAge", "jobDiploma",
# "jobDesc",
# "jobType", "jobNum", "jobWorkCity", "jobWorkLoc", "jobWelfare", "jobMajorList", "age", "gender",
# "email",
# "jobCVformat", "jobMinimumDays", "jobSkill", "jobCertificate"]
jdInc = OrderedDict()
for k in inc_keys:
jdInc[k] = ""
self.result["jdInc"] = jdInc
jdJob = OrderedDict()
# for k in job_keys:
# jdJob[k] = ""
self.result["jdJob"] = jdJob
self.first = []
self.second = []
self.third = []
# 记录jd是何种类型的，是否名企界面，是否包含表格
self.jdType = None
# 是否包含表格
self.has_table = False
# 职位名列表
self.jobNameList = []
self.jobNameLine = []
self.jobType = []
# 专业列表和技能列表
self.majorList = []
self.skillList = []
self.intro_range = []
# 用来存职位信息
self.jdJob = defaultdict(lambda: defaultdict(unicode))
def replace_space(self, line):
'''
输入：一行文本
输出：将具有多个空格的文本替换为一个空格之后的文本
'''
regex = re.compile(u' +')
line = re.sub(regex, u' ', line)
return line
def judge_eng(self, word):
'''
输入：一个单词
功能：判断这个单词是否为英文
'''
if len(re.split(u"\w", word.lower())) > 4:
return True
else:
return False
def clean_line(self, line):
"""
清除一个句子首尾的标点符号
"""
line = self.CLEAN_LINE.sub("", line).strip()
line = re.sub("\s+|^/d+[；’、，／。\.]", "", line)
return line
def clean_cnNum(self, line):
"""
经验年限提取时，中文一二三等转为123
"""
line = unicode(line)
a = [u"一", u"二", u"三", u"四", u"五", u"六", u"七", u"八", u"九", u"十", u"两"]
b = range(1, 11) + [2]
table = dict((ord(aa), bb) for aa, bb in zip(a, b))
return line.translate(table)
def line2vec(self, line):
"""
句子转换为向量
"""
vec = np.zeros(50)
for word in jieba.cut(line):
if word in self.w2v.vocab:
vec += self.w2v[word]
return vec
def clean_jobname(self, jobname):
"""
职位名清洗
"""
print jobname
if jobname.lower() in self.jobdic:
return jobname
else:
res = [(lcs_len(jobname, job), job) for job in self.jobdic]
res.sort()
return res[-1][1]
def desc_extract(self, soup):
line_list = soup.find_all("p")
return '\n'.join([line.get_text() for line in line_list])
# 去除img标签,1-7位空格,
removeImg = re.compile('<img.*?>| {1,7}| ')
# 删除超链接标签
removeAddr = re.compile('<a.*?>|</a>')
# 把换行的标签换为\n
replaceLine = re.compile('<tr>|<div>|</div>|</p>')
# 将表格制表<td>替换为\t
replaceTD = re.compile('<td>')
# 将换行符或双换行符替换为\n
replaceBR = re.compile('<br><br>|<br>')
# 将其余标签剔除
removeExtraTag = re.compile('<.*?>')
# 将多行空行删除
removeNoneLine = re.compile('\n+')
def replace(self, x):
x = re.sub(self.removeImg, "", x)
x = re.sub(self.removeAddr, "", x)
x = re.sub(self.replaceLine, "\n", x)
x = re.sub(self.replaceTD, "\t", x)
x = re.sub(self.replaceBR, "\n", x)
x = re.sub(self.removeExtraTag, "", x)
x = re.sub(self.removeNoneLine, "\n", x)
# strip()将前后多余内容删除
return x.strip()
# coding=utf-8
from tgrocery import Grocery
grocery = Grocery('sample')
train_src = [('education', '名师指导托福语法技巧：名词的复数形式'),
('education', '中国高考成绩海外认可 是“狼来了”吗？'),
('sports', '图文：法网孟菲尔斯苦战进16强 孟菲尔斯怒吼'),
('sports', '四川丹棱举行全国长距登山挑战赛 近万人参与')]
grocery.train(train_src)
#grocery.train('/home/wangjianfei/git/data/train_ch.txt')
# grocery.train('train_ch.txt')
grocery.save()
new_grocery = Grocery('sample')
new_grocery.load()
print(
new_grocery.predict(
'Abbott government spends $8 million on higher education media blitz'))
test_src = [
('education', '福建春季公务员考试报名18日截止 2月6日考试'),
('sports', '意甲首轮补赛交战记录:米兰客场8战不败国米10年连胜'),
]
print("start test..................")
#grocery.test('/home/wangjianfei/git/data/test.txt')
# grocery.train('train_ch.txt'))
# custom_grocery = Grocery('custom', custom_tokenize=list)
print(new_grocery.test(test_src))
class CoParserTop(object):
def __init__(self):
self.result = OrderedDict()
co_keys = ['incName', 'incAliasName', 'incScale', 'incType', 'incIndustry', 'incSubIndustry', 'incIntro',
'incIntroShort', 'incCity', 'incLocation', 'locationInfo', 'incZipCode', \
'incContactName', 'incContactEmail', 'incContactPhone', 'incContactQQ', 'incUrl', 'investIns',
'incStage', 'incLogo', 'incPhoto' \
, 'incLabel', 'prdInfo', 'leaderInfo', 'developInfo', \
'incWeiboName', 'incWeiboUrl', 'incWechatName', 'incWechatUrl', 'incWechatCode']
coInc = OrderedDict()
for k in co_keys:
coInc[k] = ""
self.result["coInc"] = coInc
self.CLEAN_TEXT = re.compile(
u"[^\u4e00-\u9fa5\w\d；:：;,。、%\.，/。！!@（）\r\n\(\)\-\+ － `]")
self.clf = Grocery(base_dir + "/jdclf")
self.clf.load()
self.SPLIT_LINE = re.compile(u"[\r\n；:：。！？;]|[ \s \xa0\u724b]{4,}")
self.CLEAN_LINE = re.compile(
u"^[\u2022（【\[\s\t\r\n\(\- ]?[\da-z１２３４５７８９]{1,2}[\.,。、，：:）】\]\)\s]|^[！＠＃￥％……＆×（）\(\)｛｝：“｜、－\-，。：:\.]|^[一二三四五六七八九１２３４５６７８９\d]{0,2}[\.、\s:： ]|[，；。、\s \.]$|^[\s \u2022 \uff0d \u25cf]"
)
self.CLEAN_JOBNAME = re.compile(
u"急聘|诚聘|高薪|包[食住宿餐]|.险一金|待遇|^急?招|职位编号\s?[\s\d:：]")
self.PAY = re.compile("(\d{3,}\-)?\d{3,}元")
self.SEX = re.compile(u"性别|男|女")
self.AGE = re.compile(u"\d+周?岁|年龄")
self.JOB_TAG = re.compile(u"全职|实习")
self.DEGREE = re.compile(
u"小学|初中|高中|职技|本科|研究生|硕士|博士|教授|专科|大专|中专|无要求|不限|无限")
self.MAIL = re.compile(u"\w+@[\w\.]+")
self.ZIP = re.compile(u"(\d{6})")
self.QQ = re.compile(u"\d{6,10}")
self.PHONE = re.compile(
u"1\d{10}|0\d{11}|\d{3,4}-\d{3,4}-\d{3,4}|\d{3,4}-\d{7,8}-\d{7,8}-\d{7,8}|\d{3,4}-\d{7,8}-\d{7,8}|\d{3,4}-\d{7,8}|\d{3,4}－\d{7,8}"
)
self.START_DEMAND = re.compile(
u"(任职资格|岗位要求|工作要求|任职条件|任职要求|职位要求)[：:\s】\n ]?")
self.START_DUTY = re.compile(
u"(工作内容|岗位职责|工作职责|职位描述|工作描述|职位介绍|职位职责|岗位描述)[:：\s 】\n ]")
self.START_BENEFIT = re.compile(u"(福利待遇|待遇|福利)[:：\s\n】]")
self.INC_URL = re.compile(u"(主页|网站|网址|官网).{0,3}[\w\d_/\.:\-]+")
self.DEMAND = re.compile(u"精通|熟悉|熟练|有.+经验")
self.DUTY = re.compile(u"负责|促成|为客户|安排的其.工作")
self.BENEFIT = re.compile(u".险一金|福利|晋身|休假|带薪|补助|补贴")
self.CERT = re.compile(
u"(\S{2,8}证书|CET-\d|普通话|英语|口语|.语|日文|雅思|托福|托业)(至少)?(通过)?[\d一二三四五六七八九]级[及或]?(以上)?|(英语)?CET-\d级?(以上)?|\
医学.{0,3}证|会计.{0,3}证|律师.{0,3}证|有.{1,8}证书")
self.degreedic = set([
line.strip()
for line in codecs.open(base_dir +
'/data/degrees.txt', 'rb', 'utf-8')
])
self.majordic = set([
line.strip()
for line in codecs.open(base_dir +
'/data/majordic.txt', 'rb', 'utf-8')
])
self.skilldic = set([
line.strip()
for line in codecs.open(base_dir +
'/data/skills.txt', 'rb', 'utf-8')
])
self.jobdic = set([
line.strip()
for line in codecs.open(base_dir +
'/data/jobnames.txt', 'rb', 'utf-8')
])
self.citydic = set([
line.strip()
for line in codecs.open(base_dir +
'/data/citydic.txt', 'rb', 'utf-8')
])
self.province_city = set([
line.strip()
for line in codecs.open(base_dir +
'/data/province_city.txt', 'rb', 'utf-8')
])
self.city_area = set([
line.strip()
for line in codecs.open(base_dir +
'/data/city_area.txt', 'rb', 'utf-8')
])
self.SALARY = re.compile(u'万')
jieba.load_userdict(base_dir + '/data/majordic.txt')
jieba.load_userdict(base_dir + '/data/skills.txt')
jieba.load_userdict(base_dir + '/data/firm.txt')
jieba.load_userdict(base_dir + '/data/degrees.txt')
jieba.load_userdict(base_dir + '/data/benefits.txt')
jieba.load_userdict(base_dir + '/data/citydic.txt')
jieba.load_userdict(base_dir + '/data/province_city.txt')
def refresh(self):
self.result = OrderedDict()
co_keys = ['incName', 'incAliasName', 'incScale', 'incType', 'incIndustry', 'incSubIndustry', 'incIntro',
'incIntroShort', 'incCity', 'incLocation', 'locationInfo','incZipCode', \
'incContactName', 'incContactEmail', 'incContactPhone', 'incContactQQ', 'incUrl', 'investIns',
'incStage', 'incLogo', 'incPhoto' \
, 'incLabel', 'prdInfo', 'leaderInfo','developInfo',\
'incWeiboName', 'incWeiboUrl', 'incWechatName', 'incWechatUrl', 'incWechatCode']
coInc = OrderedDict()
for k in co_keys:
coInc[k] = ""
self.result["coInc"] = coInc
#coding=utf-8
from tgrocery import Grocery
text_model = Grocery('all_no_town')
text_model.load()
#输入文本，预测出class_name和class_prob
def predict(text):
c = text_model.predict(' '.join(list(text)))
class_name = str(c)
class_prob = c.dec_values[class_name]
return class_name, class_prob
print predict(u'100')
tdic['id'].append(_id)
tdic['type'].append(_type)
tdic['contents'].append(contents)
i +=1
#train = pd.read_csv( train_file, header = 0, delimiter = "\t", quoting = 3 )
#test = pd.read_csv( test_file, header = 1, delimiter = "\t", quoting = 3 )
train = DataFrame(dic)
test = DataFrame(tdic)
#
#classfynews_instance 是模型保存路径
grocery = Grocery('classfynews_instance')
train_in = [train['contents'],train['type']]
grocery.train(train_in)
print grocery.get_load_status()
#grocery.save()
copy_grocery = Grocery('classfynews_instance')
copy_grocery.load()
#copy_grocery = grocery
test_in = [test['contents'],test['type']]
#输入类似 ['我是中国人','台北*****']
#输出 [11,12]
test_result = copy_grocery.predict(test['contents'])
print test_result.predicted_y
#test_result = copy_grocery.test(test_in)
#print test_result.show_result()
def main():
# Get market_sentiment of word from NTUSD-Fin
train_t = []
train_s = []
targetIn = {}
targetDict = dict()
with open('NTUSD-Fin/NTUSD_Fin_hashtag_v1.0.json', 'r',
encoding='utf-8') as f:
targetIn = json.load(f)
N = len(targetIn)
for i in range(N):
word = "#" + targetIn[i]['token']
targetDict[word] = targetIn[i]['market_sentiment']
sg = str(GroupValue_s(str(targetDict[word] / 3.5)))
train_s.append((sg, word))
with open('NTUSD-Fin/NTUSD_Fin_word_v1.0.json', 'r',
encoding='utf-8') as f:
targetIn = json.load(f)
N = len(targetIn)
for i in range(N):
word = targetIn[i]['token']
targetDict[word] = targetIn[i]['market_sentiment']
sg = str(GroupValue_s(str(targetDict[word] / 3.5)))
train_s.append((sg, word))
# Training File: Load data & Use tgrocery to train classification model
TrainingFile = open('training_set.json', 'r')
TrainingData = json.load(TrainingFile)
TrainingFile.close()
DataList = []
grocery_t = Grocery("tweet")
grocery_s = Grocery("snippet")
for DataElement in TrainingData:
tempt = DataManager()
tempt.insertData(DataElement)
tempt.group_t = GroupValue_t(tempt.sentiment)
tempt.group_s = GroupValue_s(tempt.sentiment)
line = re.sub("https?://[\w\-]+(\.[\w\-]+)+\S*", " ",
DataElement["tweet"])
train_t.append((str(tempt.group_t), line))
if isinstance(DataElement["snippet"], list):
for line in DataElement["snippet"]:
train_s.append((str(tempt.group_s), line))
elif DataElement["snippet"] != "":
train_s.append((str(tempt.group_s), DataElement["snippet"]))
else:
tempt.group_s = 0.0
DataList.append(tempt)
grocery_t.train(train_t + train_s)
grocery_t.save()
grocery_s.train(train_s)
grocery_s.save()
# Save training data created by WordScore() and GroupValue_*()
# Data will be uesd for LinearRegression() in BOTH.py
outfile = open('TG_train.txt', 'w', encoding='utf-8')
dataScore = []
dataSentiment = []
for row in DataList:
dataSentiment.append([float(row.sentiment)])
a = WordScore(row.tweet, targetDict)
b = WordScore(row.snippet, targetDict)
c = row.group_t
d = row.group_s
dataScore.append([a, b, c, d])
print(a, b, c, d, file=outfile)
outfile.close()
'''
# Train linear regression model
model = LinearRegression()
model.fit(dataScore, dataSentiment)
# Test for training data
print('(train)R-squared: %.3f' % model.score(dataScore, dataSentiment)) #0.915
predictions = model.predict(dataScore)
rms = mean_squared_error(dataSentiment,predictions)
print('RMSE: %.3f' % sqrt(rms)) #0.110
print('MSE: %.3f' % rms) #0.012
'''
# Testing File: Load data & Use tgrocery classification model to predict
TestingFile = open('test_set.json', 'r')
TestingData = json.load(TestingFile)
TestingFile.close()
DataList = []
new_grocery_t = Grocery('tweet')
new_grocery_t.load()
new_grocery_s = Grocery('snippet')
new_grocery_s.load()
for DataElement in TestingData:
tempt = DataManager()
tempt.insertData(DataElement)
line = re.sub("https?://[\w\-]+(\.[\w\-]+)+\S*", " ",
DataElement["tweet"])
tempt.group_t = float('{0}'.format(new_grocery_t.predict(line)))
value = 0.0
if isinstance(DataElement["snippet"], list):
for line in DataElement["snippet"]:
value = value + float('{0}'.format(
new_grocery_s.predict(line)))
value = value / len(DataElement["snippet"])
elif DataElement["snippet"] != "":
value = float('{0}'.format(
new_grocery_s.predict(DataElement["snippet"])))
tempt.group_s = value
DataList.append(tempt)
# Save testing data created by WordScore() and classification prediction
# Data will be uesd for LinearRegression() in BOTH.py
outfile = open('TG_test.txt', 'w', encoding='utf-8')
dataScore = []
dataSentiment = []
for row in DataList:
dataSentiment.append([float(row.sentiment)])
a = WordScore(row.tweet, targetDict)
b = WordScore(row.snippet, targetDict)
c = row.group_t
d = row.group_s
dataScore.append([a, b, c, d])
print(a, b, c, d, file=outfile)
outfile.close()
'''
grocery = Grocery('sample')
# 训练文本可以用列表传入
train_src = [
('education', '名师指导托福语法技巧：名词的复数形式'),
('education', '中国高考成绩海外认可 是“狼来了”吗？'),
('sports', '图文：法网孟菲尔斯苦战进16强 孟菲尔斯怒吼'),
('sports', '四川丹棱举行全国长距登山挑战赛 近万人参与')
]
grocery.train(train_src)
# 也可以用文件传入（默认以tab为分隔符，也支持自定义）
#grocery.train('train_ch.txt')
# 保存模型
grocery.save()
# 加载模型（名字和保存的一样）
new_grocery = Grocery('sample')
new_grocery.load()
# 预测
new_grocery.predict('考生必读：新托福写作考试评分标准')
#education
# 测试
test_src = [
('education', '福建春季公务员考试报名18日截止 2月6日考试'),
('sports', '意甲首轮补赛交战记录:米兰客场8战不败国米10年连胜'),
]
new_grocery.test(test_src)
# 输出测试的准确率
#0.5
# 同样可支持文件传入
#new_grocery.test('test_ch.txt')
ftest = open(path2, 'w')
for line in open(path):
if random.random() < theta:
ftest.write(line)
else:
ftrain.write(line)
ftrain.close()
ftest.close()
def train(path,name):
grocery = Grocery(name)
grocery.train(path)
grocery.save()
if __name__ == "__main__":
data2tt(sys.argv[3], sys.argv[1], sys.argv[2], 0.02)
train(sys.argv[1], "music")
new_grocey = Grocery("music")
new_grocey.load()
n = 0
for line in open(sys.argv[2],"r"):
ls = line.strip().split("\t")
predict = new_grocey.predict(ls[1])
test = ls[0]
result = 0
if test == str(predict):
result = 1
n += result
print predict,test,result
print n
