影响力最大化算法——MIA算法python复现

MIA算法原文：Chen W ,  Wang C ,  Wang Y . Scalable Influence Maximization for Prevalent Viral Marketing in Large-Scale Social Networks[C]// Proceedings of the 16th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, Washington, DC, USA, July 25-28, 2010. ACM, 2010.

伪代码：

 

 可根据注释和伪代码查看python代码

import sys
import time
import heapq
import numpy as np
import math

NETWORK = []
MIIA = []
MIOA = []
nodetable = []
dijk = []
Vnumber = 0
Enumber = 0
MAX = 9999  # MAX的作用是什么：表示路径距离为不可达，一般指的是算法中的∞


def MIA(k):
    S = []
    incinf = [0] * Vnumber
    for v in range(Vnumber):
        print("MIIA("+str(v)+",θ):", MIIA[v])
        for i in range(len(MIIA[v])):
            # print("tuple:", MIIA[v][i])  # 是个二元组 (节点下标, dijk) MIIA[v][i][0]就是算法line6-9中的u
            # incinf[MIIA[v][i][0]] += getAlpha_initial(v, MIIA[v][i][0])  # 对应算法line9
            incinf[MIIA[v][i][0]] += getAlpha(v, MIIA[v][i][0], S)  # 对应算法line9
    print("incinf:",incinf)
    ###main loop
    for i in range(0, k):
        u = maxp(incinf, S)  # 对应算法line14
        for j in range(len(MIOA[u])):  # MIOA[u][j][0]就是算法line16中的v
            v_line16 = MIOA[u][j][0]
            if v_line16 in S:
                continue
            else:
                # incinf[j] -= getAlpha(MIOA[u][j][0],u,S)*(1-ap(u,S,MIOA[u][j][0]))
                for l in range(len(MIIA[v_line16])):  # MIIA[v_line16][l][0]就是算法line18中的w
                    w_line18 = MIIA[v_line16][l][0]
                    if w_line18 in S:
                        continue
                    else:
                        incinf[w_line18] -= getAlpha(v_line16,w_line18,S)*(1-ap(w_line18,S,v_line16))
        S.append(u)
        for j in range(len(MIOA[u])):
            v_line23 = MIOA[u][j][0]
            if v_line23 in S:
                continue
            else:
                # incinf[j] += getAlpha(MIOA[u][j][0],u,S)*(1-ap(u,S,MIOA[u][j][0]))
                for l in range(len(MIIA[v_line23])):
                    w_line27 = MIIA[v_line23][l][0]
                    if w_line27 in S:
                        continue
                    else:
                        incinf[w_line27] += getAlpha(v_line23, w_line27, S) * (1 - ap(w_line27, S, v_line23))
    return S

# 求incinf最大的非种子节点
def maxp(incinf, S):
    maxp = 0  # incinf最大的节点下标 max point
    maxv = 0  # 最大的incinf值      max value
    for i in range(Vnumber):
        if i in S:  # 跳过种子
            continue
        elif incinf[i] > maxv:  # 非种子节点incinf比之前的大  找最大值通用步骤
            maxv = incinf[i]
            maxp = i
    return maxp


def getAlpha(v, u, S):
    print("get α(" + str(v) + "," + str(u) + ")", S)
    if v == u:
        return 1
    else:
        print("--MIOA(" + str(u) + ",θ):", MIOA[u])
        if MIOA[u] == []:
            return 0
        w = MIOA[u][0][0]  ### 0 ???  这里看来是随便选择了u的某一个出邻居w
        print("w:", w)
        if S == []:
            return getAlpha(v, w, S) * math.pow(math.e, -MIOA[u][0][1])  # pp = e^(-dijk) <=> -log(pp) = dijk
        if w in S:
            return 0
        else:
            listV = []
            for i in range(len(MIIA[v])):
                listV.append(MIIA[v][i][0])  # listV 储存所有在MIIA(v,θ)中的节点下标
            Nin = []
            for i in range(len(MIIA[w])):
                if MIIA[w][i][0] in listV:  # MIIA[w][i][0]、Nin[i][0]对应算法2 line9的 u'
                    Nin.append(MIIA[w][i])

            pun = 1
            for i in range(len(Nin)):
                if Nin[i][0] == u:
                    continue
                else:
                    pun = pun * (1 - ap(Nin[i][0], S, v) * math.pow(math.e, -Nin[i][1]))  # pp = e^(-dijk) <=> -log(pp) = dijk
            return getAlpha(v, w, S) * math.pow(math.e, -MIOA[u][0][1]) * (1 - pun)




def getAlpha_initial(v, u):
    print("get α("+str(v)+","+str(u)+")")
    if v == u:
        return 1
    else:
        print("--MIOA(" + str(u) + ",θ):", MIOA[u])
        if MIOA[u] == []:
            return 0
        w = MIOA[u][0][0]  ### 0 ???  这里看来是随便选择了u的某一个出邻居w
        print("w:",w)
        # return getAlpha_initial(v, w) * MIOA[u][0][1]
        return getAlpha_initial(v, w) * math.pow(math.e, -MIOA[u][0][1])  # pp = e^(-dijk) <=> -log(pp) = dijk


def ap(u, S, v):
    print("get ap("+str(u)+","+str(v)+")", S)
    listV = []
    for i in range(len(MIIA[v])):
        listV.append(MIIA[v][i][0])  # listV 储存所有在MIIA(v,θ)中的节点下标
    Nin = []
    for i in range(len(MIIA[u])):
        if MIIA[u][i][0] in listV:  # MIIA[u][i][0]、Nin[i][0]对应算法2 line6的 w
            Nin.append(MIIA[u][i])  # Nin 储存的是既在MIIA(v,θ)又在MIIA(u,θ)中的节点w 的 (节点下标, dijk(w,u))二元组

    if u in S:
        return 1
    elif Nin == []:
        return 0
    else:
        pun = 1
        for i in range(len(Nin)):  #
            # pun = pun * (1 - ap(Nin[i][0], S, v) * Nin[i][1])  ##CUN YI !!!  ap(Nin[i][0], S, v)
            pun = pun * (1 - ap(Nin[i][0], S, v) * math.pow(math.e, -Nin[i][1]))  # pp = e^(-dijk) <=> -log(pp) = dijk
            print("pp("+str(Nin[i][0])+","+str(u)+")")
        return 1 - pun


def Load_data_IC(network_file):
    global NETWORK, nodetable, Vnumber, Enumber, MIIA, MIOA
    file = open(network_file, "r", encoding="utf-8")
    Vnumber, Enumber = file.readline().split()
    Vnumber = int(Vnumber)
    Enumber = int(Enumber)
    nodetable = [[MAX for j in range(Vnumber)] for j in range(Vnumber)]
    # (weight, dijk)
    for i in range(0, Vnumber):
        NETWORK.append([])
        MIIA.append([])
        MIOA.append([])
    for i in range(0, Enumber):
        start, end, weight = file.readline().split()
        # start = int(start)
        # end = int(end)
        start = int(start)-1  # 数据集中节点id从1开始 故此处减1 对应着list的下标从0开始 节点id=节点下标+1
        end = int(end)-1      # 数据集中节点id从1开始 故此处减1 对应着list的下标从0开始 节点id=节点下标+1
        weight = float(weight)
        NETWORK[start].append([end, weight])
        # nodetable[start][end] = math.log(weight)
        nodetable[start][end] = -math.log(weight)  # 根据MIA论文 应对权重取-loge 这样最短路径距离之和最小<=>pp值最大
    return nodetable
    # 行数 = v
    # 每一行[:]，[end, weight]


def getdijk_IC():
    global dijk
    for i in range(0, Vnumber):
        ##print(i)
        dijk.append(Dijkstra(nodetable, i))
    return dijk


def Dijkstra(table, node):
    final = [0] * (Vnumber)
    distance = [0] * (Vnumber)
    path = [0] * (Vnumber)

    print("node:" + str(node),"-"*15)
    for i in range(0, Vnumber):
        distance[i] = table[node][i]
        if distance[i] != MAX:
            path[i] = node
        else:
            path[i] = MAX
    print("distance:", distance)
    print("path:", path)
    final[node] = 1
    path[node] = node
    print("final:", final)
    print("path:", path)
    k = 0
    for i in range(0, Vnumber):
        min = MAX
        for j in range(0, Vnumber):
            if distance[j] < min and final[j] == 0:
                min = distance[j]
                k = j
        final[k] = True
        for j in range(0, Vnumber):
            if (distance[j] > min + table[k][j]) and final[j] == 0:
                distance[j] = min + table[k][j]
                path[j] = k
    distance[node] = 0
    print("*"*17+"final distance of",node,":", distance)
    return distance


def getMMM(theta):
    global MIIA, MIOA
    for i in range(Vnumber):
        for j in range(Vnumber):
            # if dijk[i][j] > 1 or i == j:
            if dijk[i][j] == MAX or i == j:  # 不可达 或者 是自己，则不加入自己的MIIA或MIOA领域内
                continue
            # elif dijk[i][j] >= theta:
            elif dijk[i][j] <= -math.log(theta):  # dijk = -log(pp), pp<=theta <=> -log(pp)>=-log(theta)
                MIIA[j].append((i, dijk[i][j]))
                MIOA[i].append((j, dijk[i][j]))
    print("MIIA:", MIIA)
    print("MIOA:", MIOA)


if __name__ == '__main__':
    # python IMP.py –i network.txt -k 5 -m IC -t 60
    time_start = time.time()
    #network_file = sys.argv[2]
    #seed_size = int(sys.argv[4])
    #diffusionmodel = sys.argv[6]
    #TIME = int(sys.argv[8])
    # Amazon2 DBLP2 network network5 testNetwork
    # seeds seeds0 seeds2种子文件  Amazon DBLP只有两列
    network_file = "F:/py/Network.txt"  # network  testNetwork
    nodetable = Load_data_IC(network_file)
    print(nodetable)

    print("begindijk", time.time() - time_start)
    dijk = getdijk_IC()
    getMMM(1.928749847963923e-22)  # 1.928749847963923e-22是e^-50  0.05
    # print(MIOA[10])  # 打印节点11的最大影响出度节点及其最短路径距离

    print("beginmia", time.time() - time_start)
    S = MIA(5)
    time_end = time.time()
    print(S)
    print(time_end - time_start)
#[33, 51, 60, 46, 53]

# 改Load_data_IC [26, 27, 30, 42, 46]

# 改getMMM [32, 43, 58, 60, 29]