def train_paper():
things = [p["in_citations"] for p in col_paper.find()]
things = flatten(things, dtype="s")
things = list(set(things))
print(len(things))
p = [y['url_id'] for y in col_paper.find()]
# 剔除已经存在的部分
things = [x for x in things if x not in p]
# things = [x['_id'] for x in col_page.find({"used": False})]
au = PaperPageThread(things, 56)
s = time()
au.start()
e = time()
print("The time spent on this program is %f s" % (e - s))
def ar_train_data():
process = 0
all = col_paper.count()
papers = [x['url_id'] for x in col_paper.find()]
with open("../data_set/ar_train.data", "w", encoding='utf-8') as f:
for p in col_paper.find(no_cursor_timeout=True):
authors = p['authors']
r_authors = []
in_citations = p['in_citations']
in_citations = [x for x in in_citations if x in papers]
for oc in in_citations:
oc_paper = col_paper.find_one({'url_id': oc})
r_authors += oc_paper['authors']
r_authors = list(set(r_authors))
sentence = " ".join(enlarge_author(authors, r_authors))
f.write(sentence + '\n')
f.flush()
process += 1
print("\rprocessing...\t%.2f%%" % (process / all * 100), end="")
def author():
things = []
for p in col_paper.find():
things += p['authors']
things = list(set(things))
au = AuthorPageThread(things, thread_num=56)
s = time()
au.start()
e = time()
print("The time spent on this program is %f s" % (e - s))
def idx_paper():
"""
重新对数据库中的所有文档赋予id值
:return:
"""
i = 1
for au in col_paper.find():
col_author.update_one({'_id': au['_id']}, {"$set": {'id': i}})
print("\r正在处理：", i, end='')
i += 1
print("\n完成id重新赋值")
def init_data_net(self):
# 从远程数据库中将文档的标题以及引文纳入训练
for p in col_paper.find():
sentence = segment(p['abstract']) + segment(p['title'])
if sentence:
self.data.append(sentence)
self.data_label.append(p['url_id'])
with open("CR/CR_data_lable.bin", 'wb') as f:
# 将系统数据进行本地缓存，读取时可用data,data_label=pickle.load(open("CR_data_lable.bin", 'wb'))
pickle.dump([self.data, self.data_label], f)
self.paper_count = len(self.data_label)
def pr_train_data():
process = 0
all = col_paper.count()
with open("../data_set/pr_train.data", "w", encoding='utf-8') as f:
for p in col_paper.find():
url_id = [p['url_id']]
in_citations = p['in_citations']
sentence = " ".join(enlarge_author(url_id, in_citations))
f.write(sentence + '\n')
f.flush()
process += 1
print("\rprocessing...\t%.2f%%" % (process / all * 100), end="")
def get_ar_authors(filter: {}):
"""
根据相关条件查询文档中所有相关作者信息
:param filter:
:return:
"""
rs = {}
papers = [x['url_id'] for x in col_paper.find()]
for p in col_paper.find(filter, no_cursor_timeout=True):
authors = p['authors']
r_authors = []
in_citations = p['in_citations']
in_citations = [x for x in in_citations if x in papers]
for oc in in_citations:
oc_paper = col_paper.find_one({'url_id': oc})
r_authors += oc_paper['authors']
r_authors = list(set(r_authors))
ar_authors = list(set(r_authors + authors))
rs[p['url_id']] = [authors, r_authors, ar_authors]
if filter == {}:
with open("../data_set/ar_author.json", "w", encoding='utf-8') as f:
json.dump(rs, f)
return rs
def test(self):
test_paper = [x['url_id'] for x in col_testPaper.find()]
in_citations = [x['in_citations'] for x in col_paper.find()]
total = 0
true = 0
for t in range(len(test_paper)):
mode_sim = self.model.wv.most_similar(test_paper[t],
topn=self.topn)
call = [x[0] for x in mode_sim]
inc = [x for x in in_citations[t] if x in self.sys_docs]
total += len(inc)
true = len([x for x in call if x in inc])
r_rate = float(true / total)
t_rate = float(true / (len(test_paper) * self.topn))
print("召回率：%.2f%%" % r_rate)
print("准确率：%.2f%%" % t_rate)
def test(self):
test_paper = [x['url_id'] for x in col_testPaper.find()]
in_citations = [x['in_citations'] for x in col_paper.find()]
true = 0
total = 0
for i in range(len(test_paper)):
t_vec = sent2vec(self.model,
self.data[self.data_label.index(test_paper[i])])
call = self.most_sim(t_vec)
inc = [x for x in in_citations[i] if x in self.sys_docs]
inc_index = [self.data_label.index(x) for x in inc]
total += len(inc)
true += len([x for x in inc_index if x in call])
r_rate = float(true / total)
t_rate = float(true / (len(test_paper) * self.topn))
print("召回率：%.2f%%" % r_rate)
print("准确率：%.2f%%" % t_rate)
class CR:
data = []
data_label = []
paper_count = 0
model = None
sims = None # 目标文档与系统文档的相关性大小
topn = 100 # 返回的结果数
sort_sim = None # 相关性排序结果，只有文档id
sys_docs = [x["url_id"] for x in col_paper.find()]
sys_docs_cr = []
def __init__(self, topn=100, doc=None):
"""
:param topn:返回相关结果数
:param doc: 目标文档，包含标题与摘要组成的字符串
"""
self.topn = topn
if doc:
self.aim_doc = segment(doc)
self.aim_doc_vec = sent2vec(words=self.aim_doc)
def init_data_net(self):
# 从远程数据库中将文档的标题以及引文纳入训练
for p in col_paper.find():
sentence = segment(p['abstract']) + segment(p['title'])
if sentence:
self.data.append(sentence)
self.data_label.append(p['url_id'])
with open("CR/CR_data_lable.bin", 'wb') as f:
# 将系统数据进行本地缓存，读取时可用data,data_label=pickle.load(open("CR_data_lable.bin", 'wb'))
pickle.dump([self.data, self.data_label], f)
self.paper_count = len(self.data_label)
def init_data_local(self):
"""
当本地已缓存有数据时，直接读取
:return:
"""
self.data, self.data_label = pickle.load(
open("CR/CR_data_lable.bin", 'rb'))
def get_sys_docs_cr(self):
if self.sys_docs_cr:
return self.sys_docs_cr
for d in self.data:
self.sys_docs_cr.append(sent2vec(self.model, d))
return self.sys_docs_cr
def get_model(self):
if not self.model:
self.model = Doc2Vec.load("CR/CR.model")
return self.model
def train(self):
sentences = LabeledLineSentence(self.data, self.data_label)
# 迭代20次
self.model = Doc2Vec(size=100,
window=10,
min_count=3,
workers=10,
iter=20)
self.model.build_vocab(sentences)
print("开始训练...")
# 训练模型
start = time.time()
self.model.train(sentences,
total_examples=self.model.corpus_count,
epochs=12)
self.model.save("CR/CR.model")
self.model.save_word2vec_format("CR/CR.vector")
print("模型已保存，共花时间：", time.time() - start)
def most_sim(self, doc_vec):
# 建立相关性索引
sims = sim(doc_vec, self.get_sys_docs_cr())
# 对相关度进行排序
call = np.argsort(-sims)
return call[:self.topn]
def test(self):
test_paper = [x['url_id'] for x in col_testPaper.find()]
in_citations = [x['in_citations'] for x in col_paper.find()]
true = 0
total = 0
for i in range(len(test_paper)):
t_vec = sent2vec(self.model,
self.data[self.data_label.index(test_paper[i])])
call = self.most_sim(t_vec)
inc = [x for x in in_citations[i] if x in self.sys_docs]
inc_index = [self.data_label.index(x) for x in inc]
total += len(inc)
true += len([x for x in inc_index if x in call])
r_rate = float(true / total)
t_rate = float(true / (len(test_paper) * self.topn))
print("召回率：%.2f%%" % r_rate)
print("准确率：%.2f%%" % t_rate)
class PR:
"""
输出模型文件：
PR.model：模型参数
PR.vector:词向量
"""
data_path = "../data_set/pr_train.data"
model_path = "PR/PR.model" # 模型参数缓存文件路径
vector_path = "PR/PR.vector" # 词向量缓存路径
model = None
sims = None # 目标文档与系统文档的相关性大小
topn = 100 # 返回的结果数
sort_sim = None # 相关性排序结果，只有文档idy
sys_docs = [x["url_id"] for x in col_paper.find()]
def __init__(self, topn=100):
self.topn = topn
def init_data_net(self):
"""
生成ar.text文件
:return:
"""
pass
def train(self):
start = time.time()
# 直接从文本文件中读取数据集
sentences = LineSentence(self.data_path)
self.model = Word2Vec(sentences,
min_count=2,
sg=1,
hs=1,
size=100,
window=5,
workers=4,
iter=20)
self.model.save(self.model_path)
self.model.wv.save_word2vec_format(self.vector_path, binary=False)
print("模型已保存，训练共花时间：", time.time() - start)
def test(self):
test_paper = [x['url_id'] for x in col_testPaper.find()]
in_citations = [x['in_citations'] for x in col_paper.find()]
total = 0
true = 0
for t in range(len(test_paper)):
mode_sim = self.model.wv.most_similar(test_paper[t],
topn=self.topn)
call = [x[0] for x in mode_sim]
inc = [x for x in in_citations[t] if x in self.sys_docs]
total += len(inc)
true = len([x for x in call if x in inc])
r_rate = float(true / total)
t_rate = float(true / (len(test_paper) * self.topn))
print("召回率：%.2f%%" % r_rate)
print("准确率：%.2f%%" % t_rate)
def get_model(self):
if not self.model:
self.model = Word2Vec.load("PR.model")
return self.model
