import math
import random
import easygraph as eg
import numpy as np
[docs]
class NodeParams:
def __init__(self, active, inWeight, threshold):
self.active = active
self.inWeight = inWeight
self.threshold = threshold
[docs]
def structural_hole_influence_index(
G_original,
S,
C,
model,
variant=False,
seedRatio=0.05,
randSeedIter=10,
countIterations=100,
Directed=True,
):
"""Returns the SHII metric of each seed.
Parameters
----------
G_original: easygraph.Graph or easygraph.DiGraph
S: list of int
A list of nodes which are structural hole spanners.
C: list of list
Each list includes the nodes in one community.
model: string
Propagation Model. Should be IC or LT.
variant: bool, default is False
Whether returns variant SHII ml_metrics or not.
variant SHII = # of the influenced outsider / # of the influenced insiders
SHII = # of the influenced outsiders / # of the total influenced nodes
seedRatio: float, default is 0.05
# of sampled seeds / # of nodes of the community that the given SHS belongs to.
randSeedIter: int, default is 10
How many iterations to sample seeds.
countIterations: int default is 100
Number of monte carlo simulations to be used.
Directed: bool, default is True
Whether the graph is directed or not.
Returns
-------
seed_shii_pair : dict
the SHII metric of each seed
Examples
--------
# >>> structural_hole_influence_index(G, [3, 20, 9], Com, 'LT', seedRatio=0.1, Directed=False)
References
----------
.. [1] https://dl.acm.org/doi/pdf/10.1145/2939672.2939807
.. [2] https://github.com/LifangHe/KDD16_HAM/tree/master/SHII_metric
"""
if not Directed:
G = eg.DiGraph()
for edge in G_original.edges:
G.add_edge(edge[0], edge[1])
G.add_edge(edge[1], edge[0])
else:
G = G_original.copy()
# form pair like {node_1:community_label_1,node_2:community_label_2}
node_label_pair = {}
for community_label in range(len(C)):
for node_i in range(len(C[community_label])):
node_label_pair[C[community_label][node_i]] = community_label
# print(node_label_pair)
seed_shii_pair = {}
for community_label in range(len(C)):
nodesInCommunity = []
seedSetInCommunity = []
for node in node_label_pair.keys():
if node_label_pair[node] == community_label:
nodesInCommunity.append(node)
if node in S:
seedSetInCommunity.append(node)
seedSetSize = int(math.ceil(len(nodesInCommunity) * seedRatio))
if len(seedSetInCommunity) == 0:
continue
for seed in seedSetInCommunity:
print(">>>>>> processing seed ", seed, " now.")
oneSeedSet = []
if node not in oneSeedSet:
oneSeedSet.append(seed)
seedNeighborSet = []
# using BFS to add neighbors of the SH spanner to the seedNeighborSet as seed candidates
queue = []
queue.append(seed)
while len(queue) > 0:
cur_node = queue[0]
count_neighbor = 0
for neighbor in G.neighbors(node=cur_node):
if neighbor not in seedNeighborSet:
seedNeighborSet.append(neighbor)
count_neighbor = count_neighbor + 1
if count_neighbor > 0:
if (
len(queue) == 1
and len(oneSeedSet) + len(seedNeighborSet) < seedSetSize
):
for node in seedNeighborSet:
if node not in oneSeedSet:
oneSeedSet.append(node)
queue.append(node)
seedNeighborSet.clear()
queue.pop(0)
avg_censor_score_1 = 0.0
avg_censor_score_2 = 0.0
for randIter in range(randSeedIter):
if randIter % 5 == 0:
print("seed ", seed, ": ", randIter, " in ", randSeedIter)
randSeedSet = []
for node in oneSeedSet:
randSeedSet.append(node)
seedNeighbors = []
for node in seedNeighborSet:
seedNeighbors.append(node)
while len(seedNeighbors) > 0 and len(randSeedSet) < seedSetSize:
r = random.randint(0, len(seedNeighbors) - 1)
if seedNeighbors[r] not in randSeedSet:
randSeedSet.append(seedNeighbors[r])
seedNeighbors.pop(r)
if model == "IC":
censor_score_1, censor_score_2 = _independent_cascade(
G,
randSeedSet,
community_label,
countIterations,
node_label_pair,
)
elif model == "LT":
censor_score_1, censor_score_2 = _linear_threshold(
G,
randSeedSet,
community_label,
countIterations,
node_label_pair,
)
avg_censor_score_1 += censor_score_1 / randSeedIter
avg_censor_score_2 += censor_score_2 / randSeedIter
# print("seed ", seed, " avg_censor_score in ", randIter, "is ", censor_score_1 / randSeedIter)
if variant:
seed_shii_pair[seed] = avg_censor_score_2
else:
seed_shii_pair[seed] = avg_censor_score_1
return seed_shii_pair
def _independent_cascade(G, S, community_label, countIterations, node_label_pair):
avg_result_1 = 0
avg_result_2 = 0
N = G.number_of_nodes()
for b in range(countIterations):
# print(b, " in ", countIterations)
p_vw = np.zeros((N, N)) # 节点被激活时，激活其它节点的概率,a对b的影响等于b对a的影响
for random_i in range(N):
for random_j in range(random_i + 1, N):
num = random.random()
p_vw[random_i][random_j] = num
p_vw[random_j][random_i] = num
Q = []
activeNodes = []
for v in S:
Q.append(v)
activeNodes.append(v)
while len(Q) > 0:
v = Q[0]
for neighbor in G.neighbors(node=v):
if neighbor not in activeNodes:
toss = random.random() + 0.1
if v <= 0 or neighbor <= 0:
print(v, neighbor)
# if toss>0.5:
# activeNodes.append(neighbor)
# Q.append(neighbor)
if toss >= p_vw[v - 1][neighbor - 1]:
activeNodes.append(neighbor)
Q.append(neighbor)
Q.pop(0)
self_cov = 0
total_cov = 0
uniqueActiveNodes = []
for i in activeNodes:
if i not in uniqueActiveNodes:
uniqueActiveNodes.append(i)
for v in uniqueActiveNodes:
total_cov += 1
if node_label_pair[v] == community_label:
self_cov += 1
censor_score_1 = (total_cov - self_cov) / total_cov
censor_score_2 = (total_cov - self_cov) / self_cov
avg_result_1 += censor_score_1 / countIterations
avg_result_2 += censor_score_2 / countIterations
return avg_result_1, avg_result_2
def _linear_threshold(G, S, community_label, countIterations, node_label_pair):
tol = 0.00001
avg_result_1 = 0
avg_result_2 = 0
for b in range(countIterations):
activeNodes = []
# T is the set of nodes that are to be processed
T = []
Q = {}
for v in S:
activeNodes.append(v)
for neighbor in G.neighbors(node=v):
if neighbor not in S:
weight_degree = 1.0 / float(G.in_degree()[neighbor])
if neighbor not in Q.keys():
np = NodeParams(False, weight_degree, random.random())
Q[neighbor] = np
T.append(neighbor)
else:
Q[neighbor].inWeight += weight_degree
while len(T) > 0:
u = T[0]
if Q[u].inWeight >= Q[u].threshold + tol and not Q[u].active:
activeNodes.append(u)
Q[u].active = True
for neighbor in G.neighbors(node=u):
if neighbor in S:
continue
weight_degree = 1.0 / float(G.in_degree()[neighbor])
if neighbor not in Q.keys():
np = NodeParams(False, weight_degree, random.random())
Q[neighbor] = np
T.append(neighbor)
else:
if not Q[neighbor].active:
T.append(neighbor)
Q[neighbor].inWeight += weight_degree
if Q[neighbor].inWeight - 1 > tol:
print("Error: the inweight for a node is > 1.")
T.pop(0)
T.clear()
Q.clear()
self_cov = 0
total_cov = 0
uniqueActiveNodes = []
for i in activeNodes:
if i not in uniqueActiveNodes:
uniqueActiveNodes.append(i)
for v in uniqueActiveNodes:
total_cov += 1
if node_label_pair[v] == community_label:
self_cov += 1
censor_score_1 = (total_cov - self_cov) / total_cov # ==> SHII
censor_score_2 = (total_cov - self_cov) / self_cov
avg_result_1 += censor_score_1 / countIterations
avg_result_2 += censor_score_2 / countIterations
return avg_result_1, avg_result_2
if __name__ == "__main__":
G = eg.datasets.get_graph_karateclub()
Com = []
t1 = [1, 2, 3, 4, 5, 6, 7, 8, 11, 12, 13, 14, 17, 18, 20, 22]
Com.append(t1)
t2 = [9, 10, 15, 16, 19, 21, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34]
Com.append(t2)
print("community_label:", Com)
result = structural_hole_influence_index(
G, [3, 20, 9], Com, "IC", seedRatio=0.1, Directed=False
)
print(result)