As a python novice, my instructor asked me to use python to implement the algorithm in the paper. I was confused about the technical points required and how to implement the algorithm. Currently, I have finished the Python tutorial by Teacher Liao and am currently reading the networkx documentation.
I hope you can help me solve the following problems:
1. Technical points required to implement the algorithm
2. How to deal with this type of paper
3. Suggestions on learning in the direction of data mining
Paper address: http://cjc.ict.ac.cn/online/o...
黄舟2017-05-18 11:00:59
After a week, it has been initially completed. The extra code is not beautiful enough and inefficient. Please give me some advice
# _*_ coding:utf-8 _*_
# ==================================================================================
#
# Description: Influence Maximization on Multiple Social Networks
#
# ==================================================================================
import matplotlib.pyplot as plt
import networkx as nx
import heapq
#总图
G = nx.DiGraph()
def load_graph(file):
'''
加载文件为列表格式,并得到G,画出图结构
'''
#将总列表设成全局格式
global gllist
#迭代文件中每个元素
with open(file) as f:
lines = f.readlines()
mylist = [line.strip().split() for line in lines]
gllist = []
#将字符串型转换为整型
for i in mylist:
gllist.append(i[:-2]+map(lambda x: float(x), i[-2:]))
print '初始全局列表:'
print gllist
drawlist=[]
#提取二维列表mylist每行前三个元素,赋给新的列表drawlist
for i in range(len(mylist)):
drawlist.append([])
for j in range(3):
drawlist[i].append(mylist[i][j])
#将列表drawlist加载为有向加权图
G.add_weighted_edges_from(drawlist)
nx.draw(G, with_labels=True, width=1, node_color='y', edge_color='b')
plt.show()
print 'G图中所有节点:',G.nodes()
print 'G图中所有边:',G.edges()
print '\n'
def get_self_node(gllist, target=None):
'''
获取目标节点的自传播节点,返回selflist并包含目标节点
'''
#初始化自传播节点列表
selflist = [target]
#存放已传播节点列表
haslist = []
flag = 0
while (flag != 0):
flag = 0
for target in selflist:
if target not in haslist:
for i in range(len(gllist)):
#判断二维列表中,每行第三个元素是否为1,若为1，则为自传播节点
if ((gllist[i][0] == target)or(gllist[i][1]==target))and(gllist[i][3]==1.0):
if gllist[i][0] == target:
if gllist[i][1] not in haslist:
selflist.append(gllist[i][1])
haslist.append(gllist[i][1])
flag += 1
else:
if gllist[i][0] not in haslist:
selflist.append(gllist[i][0])
haslist.append(gllist[i][0])
flag += 1
#去除重复元素
haslist = set(haslist)
selflist = set(selflist)
#去除重复元素
selflist = set(selflist)
return selflist
def longest_path(gllist,source=None,target=None):
'''
获取起始点到实体的最大路径集合,返回为longestpath列表
'''
longestpath = []
newlist = []
for i in range(len(gllist)):
newlist.append([])
for j in range(3):
newlist[i].append(gllist[i][j])
#构建图结构
G1 = nx.DiGraph()
#添加带权有向边
G1.add_weighted_edges_from(newlist)
#获取目标节点的所有自传播街边,并存入selflist中
selflist = get_self_node(gllist, target)
max_path = 0
val_path = 1
#获取初始节点到目标节点及目标节点的自传播节点的最大路径
for v in selflist:
if v != source:
#遍历两点之间所有路径,并进行比对
for path in nx.all_simple_paths(G1,source=source,target=v):
#判断路径后两个元素是否为相同实体（如：b1->b2）
if is_self_transmit_node(path[-2], v) == 0:
for i in range(0, len(path)-1):
val_path *= G1.get_edge_data(path[i], path[i+1])['weight']
if max_path < val_path:
max_path = val_path
val_path = 1
#若目标节点为起始节点则直接跳出
else: continue ############ 有待商榷 ##############
longestpath.append(max_path)
#返回初始节点到实体的最大路径
return longestpath
def is_self_transmit_node(u, v):
'''
判断目标节点不为起始节点的自传播点
'''
flag = 0
#获得起始节点的所有自传播点
selflist = get_self_node(gllist, v)
for x in selflist:
if u == x:
flag = 1
return flag
def single_strong_infl(longestpath):
'''
计算起始点到实体的传播概率（影响强度）,返回影响强度stronginfl
'''
temp = 1
for x in longestpath:
temp *= 1-x
stronginfl = 1-temp
return stronginfl
def all_strong_infl(G):
'''
获得每个节点对实体的影响概率
'''
allstrong = [] #初始化所有节点的加权影响范围列表
gnodes = [] #初始化节点列表
tempnodes = [] #初始化临时节点列表
gnodes = G.nodes()
for u in gnodes:
strong = 0 #存储初始节点对每个实体的影响范围加权，初始化为0
#重置临时节点列表
tempnodes = G.nodes()
for v in tempnodes:
#非自身节点
if u != v:
#判断目标节点不为起始节点的自传播点
if is_self_transmit_node(v, u) == 0:
#获取起始节点到实体间最大加权路径，并存入longestpath
longestpath = longest_path(gllist, u, v)
#去除已遍历目标节点的所有自传播节点
renode = get_self_node(gllist, v)
for x in renode:
if x != v:
tempnodes.remove(x)
#计算起始节点到实体间传播概率（影响强度）
stronginfl = single_strong_infl(longestpath)
strong += stronginfl
#添加单个节点到所有实体的加权影响范围
allstrong.append([u, round(strong, 2)])
#返回每个节点到所有实体的加权影响范围
return allstrong
#output allstrong : [['a1', 2.48], ['a2', 1.6880000000000002], ['b1', 0.7], ['b2', 0], ['c1', 0], ['d2', 0.6]]
def uS_e_uppergain(u, ev, S):
'''
获取节点u在集合S的基础上对实体ev的影响增益, 传入候选节点,上界gain(u|S, ev)
'''
#获取目前实体的所有自传播节点
selflist = get_self_node(gllist, ev)
stronglist = []
#遍历自传遍节点
for v in selflist:
'''
判断节点v是否存在种子集合S中
其中v为单个节点,如v(ev, Gi)
S为种子节点集合,如['a1','a2','b1','b2','c1','d2']
'''
if v in S:
ppSv = 1
else:
longestpath = []
#遍历种子集合
for s in S:
#初始化路径权值与最大路径权值
val_path = 1
max_path = 0
#遍历两点之间所有路径，并进行比对
for path in nx.all_simple_paths(G,source=s,target=v):
#判断路径后两个元素是否为相同实体（如：b1->b2）
if is_self_transmit_node(path[-2], v) == 0:
for i in range(0, len(path)-1):
val_path *= G.get_edge_data(path[i], path[i+1])['weight']
if max_path < val_path:
max_path = val_path
#重置路径权值为1
val_path = 1
#将最大加权路径存入longestpath列表
longestpath.append(max_path)
#得到上界pp(S,v)的影响概率,上界pp(S,v)
ppSv = single_strong_infl(longestpath)
stronglist.append(ppSv)
#得到上界pp(S,ev)的影响概率,上界pp(S,ev)
ppSev = single_strong_infl(stronglist)
#获取pp(u,ev)
ppuev = single_strong_infl(longest_path(gllist, u, ev))
#计算上界gain(u|S，ev)
uSevgain = (1 - ppSev) * ppuev
return uSevgain
def uppergain(u, emu, ems, S):
'''
在已有种子集合S的基础上,求得节点u的影响增益上界,
其中传进参数ems为二维列表,如[['a1',2.48],['a2',1.688]],S则为['a1','a2']
'''
uSgain = 0.0
#遍历emu得到列表形式,得到如['a1',2.48]形式
for ev in emu:
#判断节点是否存在种子集合中
if ev[0] in S:
uSgain += uS_e_uppergain(u, ev[0], S)
else:
uSgain += ev[1]
#返回上界gain(u|S)
return uSgain
def bound_base_imms(G, k):
'''
完全使用影响增益上界的方式选择top-k个种子节点的过程
'''
#初始化emu,H,初始化ems=空集,S=空集
Htemp = []
Htemp = all_strong_infl(G)
H = []
#遍历Htemp=[['a1',2.48],['a2',1.688]],得到如['a1',2.48]形式
for x in Htemp:
#逐个获取二维列表中每一行，形式为['a1',2.48,0]
H.append([x[0],x[1],0])
emu = []
emu = all_strong_infl(G)
ems = []
S = []
for i in range(k):
#提取堆顶元素,tnode的形式为['a1',2.48,0]
tnode = heapq.nlargest(1, H, key=lambda x: x[1])
#将[['b2', 3.1, 0]]格式改为['b2', 3.1, 0]格式
tnode = sum(tnode, [])
while (tnode[2] != i):
gain = 0.0
#获取节点u的影响增益上界
gain = uppergain(tnode, emu, ems, S)
#赋值影响范围
tnode[1] = gain
#修改status
tnode[2] = i
#对堆进行排序
H = heapq.nlargest(len(H), H, key=lambda x: x[1])
#获取堆顶元素
tnode = heapq.nlargest(1, H, key=lambda x: x[1])
tnode = sum(tnode, [])
#添加node到种子集合
S.append([tnode[0]])
#更新ems,添加新节点及节点对每个实体的影响范围加权
ems.append([tnode[0], tnode[1]])
#删除堆顶元素
H.remove(tnode)
print ems
return sum(S, [])
if __name__=='__main__':
#大小为k的种子集合S
k = 60
#加载文件数据,得到图G和初始列表gllist
load_graph('test.txt')
#完全使用影响增益上界值的计算过程函数,打印种子集合S
print '种子集合:',bound_base_imms(G, k)test.txt
a1 b1 0.2 0
a1 c1 0.8 0
a2 b2 0.4 0
a2 d2 1 0
b1 c1 0.7 0
c2 a2 0.8 0
d2 b2 0.6 0
a1 a2 1 1
a2 a1 0.1 1
....
a1 l1 0.5 0
a1 m1 0.5 0
a1 q1 0.5 0
a1 v1 0.5 0
a1 z1 0.5 0
a1 s1 0.5 0
a1 w1 0.5 0
a1 u1 0.5 0
the first two listed as Spreading entities, the third column represents the propagation probability between entities, the last column represents 0 for propagation in the same network, and 1 represents self-propagation between networks.
The next step is to optimize:
1. Use an independent cascade model and set the threshold
2. Change the maximum path to the shortest path and use log