-
Notifications
You must be signed in to change notification settings - Fork 12
/
extract_inf2vec_trainset.py
235 lines (196 loc) · 8.55 KB
/
extract_inf2vec_trainset.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
# -*- coding: utf-8 -*-
"""
Extract the propagation network of each cascade and the train set for Inf2vec
"""
import igraph as ig
import time
import pandas as pd
import numpy as np
from datetime import datetime
def sort_papers(papers):
"""
# Sort MAG diffusion cascade, which is a list of papers and their authors, in the order the paper'sdate
"""
x =list(map(int,list(map(lambda x:x.split()[-1],papers))))
return [papers[i].strip() for i in np.argsort(x)]
def remove_duplicates(cascade_nodes,cascade_times):
"""
# Some tweets have more then one retweets from the same person
# Keep only the first retweet of that person
"""
duplicates = set([x for x in cascade_nodes if cascade_nodes.count(x)>1])
for d in duplicates:
to_remove = [v for v,b in enumerate(cascade_nodes) if b==d][1:]
cascade_nodes= [b for v,b in enumerate(cascade_nodes) if v not in to_remove]
cascade_times= [b for v,b in enumerate(cascade_times) if v not in to_remove]
return cascade_nodes, cascade_times
def store_samples(fn,cascade_nodes,cascade_times,initiators,op_id, op_time,to_train_on,sampling_perc=120):
"""
# Store the samples for the train set as described in the node-context pair creation process for INFECTOR
"""
#---- Inverse sampling based on copying time
#cascade_nodes=cascade_nodes[1:]
#cascade_times=cascade_times[1:]
no_samples = round(len(cascade_nodes)*sampling_perc/100)
if(fn=="weibo"):
times = [1*1.0/(abs((cascade_times[i]-op_time).total_seconds())+1) for i in range(0,len(cascade_nodes))]
else:
times = [1*1.0/(abs((cascade_times[i]-op_time))+1) for i in range(0,len(cascade_nodes))]
s_times = sum(times)
if s_times==0:
samples = []
else:
probs = [float(i)/s_times for i in times]
samples = np.random.choice(a=cascade_nodes, size=int(no_samples), p=probs)
casc_len = str(len(cascade_nodes))
#----- Store train set
if(fn=="mag"):
for op_id in initiators:
for i in samples:
#---- Write inital node, copying node,length of cascade
to_train_on.write(str(op_id)+","+i+","+casc_len+"\n")
else:
for i in samples:
#if(op_id!=i):# though this can t be
#---- Write initial node, copying node, copying time, length of cascade
to_train_on.write(str(op_id)+","+i+","+casc_len+"\n")
def run_rwr(prop_net,restart = 0.5,path_size = 10):
"""
# Run a Random Walk with restart to retreive a set of nodes for each node
"""
train_set = {}
for v in prop_net.vs:
steps = 0
rwr = []
#---- RWR on v
current = v
while steps<path_size:
steps+=1
#-- Jump randomly in one of the neighbors
neighs = prop_net.neighbors(current)
if(len(neighs)==0):
continue
current = prop_net.vs[np.random.choice(neighs)]
rwr.append(current["name"])
if np.random.choice(2,p=[1-restart,restart]):
current = v
if(len(rwr)==5):
break
#----- Random sample
train_set[v] = rwr+list(np.random.choice(prop_net.vs["name"],45))
return train_set
def run(fn,sampling_perc,log):
print("Reading the network")
g = ig.Graph.Read_Ncol(fn+"/"+fn+"_network.txt")
#------ in mag it is undirected
#if fn =="mag":
# g.to_undirected()
start = time.time()
f = open(fn+"/train_cascades.txt","r")
inf2vec_set = open(fn+"/inf2vec_set.txt","a")
#----- Initialize features
idx = 0
deleted_nodes = []
g.vs["Cascades_participated"] = 0
if(fn=="mag"):
start_t = int(next(f))
idx=0
start = time.time()
#---------------------- Iterate through cascades to create the train set
for line in f:
if(fn=="mag"):
parts = line.split(";")
initiators = parts[0].replace(",","").split(" ")
op_time = int(initiators[-1])+start_t
initiators = initiators[:-1]
papers = parts[1].replace("\n","").split(":")
papers = sort_papers(papers)
papers = [list(map(lambda x: x.replace(",",""),i)) for i in list(map(lambda x:x.split(" "),papers))]
#---- Extract the authors from the paper list
flatten = []
for i in papers:
flatten = flatten+i[:-1]
u,i = np.unique(flatten,return_index=True)
cascade_nodes = list(u[np.argsort(i)])
cascade_times = []
cascade_nodes = []
for p in papers:
tim = int(p[-1])+start_t
for j in p[:-1]:
if j!="" and j!=" " and j not in cascade_nodes:
try:
#----- to check if the node is in the network
g.vs.find(name=j)["Cascades_participated"]+=1
except:
continue
cascade_nodes.append(j)
cascade_times.append(tim)
#---- Define the propagation network
prop_net = ig.Graph(directed=True)
tmpt = initiators+cascade_nodes
prop_net.add_vertices(tmpt)
#--- Draw edges between initiators and the rest
for i in initiators:
for j in cascade_nodes:
try:
#------ i is co authors with j
edge = g.get_eid(i,j)
#------ i influences j
prop_net.add_edge(i,j)
except:
continue
#--- Draw edges between cascade nodes
for p in range(0,len(papers)):
# if it takes more than 15 mins, go away)
for i in papers[p]:
#---- Draw edges to the authors from the subsequent papers
for p2 in range(p+1,len(papers)):
for j in papers[p2]:
try:
#------ i is followed by j
edge = g.get_eid(i,j)
#------ i influences j
prop_net.add_edge(i,j)
except:
continue
else:
initiators = []
cascade = line.replace("\n","").split(";")
if(fn=="weibo"):
cascade_nodes = list(map(lambda x: x.split(" ")[0],cascade[1:]))
cascade_times = list(map(lambda x: datetime.strptime(x.replace("\r","").split(" ")[1], '%Y-%m-%d-%H:%M:%S'),cascade[1:]))
else:
cascade_nodes = list(map(lambda x: x.split(" ")[0],cascade))
cascade_times = list(map(lambda x: int(x.replace("\r","").split(" ")[1]),cascade))
#---- Remove retweets by the same person in one cascade
cascade_nodes, cascade_times = remove_duplicates(cascade_nodes,cascade_times)
#---------- Dictionary nodes -> cascades
op_id = cascade_nodes[0]
op_time = cascade_times[0]
if(len(cascade_nodes)<2):
continue
#---- Derive propagation network
prop_net = ig.Graph(directed=True)
prop_net.add_vertices(cascade_nodes)
#---- Data-based weighing
for i in range(len(cascade_nodes)):
# if it takes more than 15 mins, go away)
#--- Add to the propagation graph all i's edges that point to j
for j in range(i+1,len(cascade_nodes)):
try:
#------ i is followed by j
edge = g.get_eid(cascade_nodes[j],cascade_nodes[i])
except:
continue
pairs = run_rwr(prop_net)
for node,rwr in pairs.iteritems():
for r in rwr:
inf2vec_set.write(node["name"]+","+str(r)+"\n")
idx+=1
if(idx%1000==0):
print("-------------------",idx)
idx=0
inf2vec_set.close()
print("Number of nodes not found in the graph: ",len(deleted_nodes))
f.close()
log.write("inf2vec preprocess time "+fn+" "+str(time.time()-start)+"\n")