社交网络影响力最大化 | CELF算法解决IM问题

最新推荐文章于 2020-12-18 10:26:09 发布

weixin_41668995

最新推荐文章于 2020-12-18 10:26:09 发布

阅读量817

点赞数

分类专栏： IM 文章标签：贪心算法 np问题社交网络影响力最大化

本文链接：https://blog.csdn.net/weixin_41668995/article/details/111354192

版权

IM 专栏收录该内容

2 篇文章

订阅专栏

CELF

数据集：facebook

# -*- coding:utf8 -*-
import numpy as np
import time
from collections import Counter
from igraph import *
import random
import copy

import matplotlib.pyplot as plt
from random import uniform, seed
import pandas as pd
import time
from collections import Counter



def generate_p(num):
    p = np.random.rand(num)
    return p

def IC(g, S, mc):
    print('-----v', S)
    spread = []
    for i in range(mc):
        new_nodes, A = S[:], S[:]
        while new_nodes != []:
            node_out = g.loc[g['source'].isin(new_nodes)]
            node_out = node_out['target']
            node_out = node_out.tolist()
            success = np.random.uniform(0, 1, len(node_out)) < generate_p(len(node_out))
            new_ones = list(np.extract(success, np.array(node_out)))

            new_nodes = list(set(new_ones) - set(A))
            print('new nodes', new_nodes)

            A += new_nodes

        spread.append(len(A))
        print(spread)

    return np.mean(spread)


def CELF(g, k, mc):
    g_v = g['source'].tolist()
    g_v = list(set(g_v))
    v_count = len(g_v)

    S, SPREAD, timelapse, start_time = [], [], [], time.time()

    # margin_gain = [ IC(g, [v], mc) for v in g_v ]
    # look_up = dict(zip(g_v, margin_gain))
    look_up = look
    look_up = dict(sorted(look_up.items(), key = lambda x:(x[1]), reverse=True))
    print('lookup', look_up)

    s = list(look_up.keys())[0]  # 第一个加入的种子节点
    S.append(s)  # 将第一个种子节点加入种子集
    spread = list(look_up.values())[0]  # 第一个种子的影响力传播值
    print('spread', spread)
    SPREAD.append(list(look_up.values())[0])  # 将第一个种子的影响力传播加入影响力传播表

    spread_max, node_max = -1, -1

    for s_index in range(k-1):
        flag = 0
        for i_index, i in enumerate(look_up.keys()):
            print('i_index, i', i_index, i)
            print(look_up)
            if i not in S:
                spread_i = IC(g, S + [i], mc)
                spread_i_margin = spread_i - spread

                spread_next_i = i_index + 1 if i_index+1 < v_count else i_index

                if spread_i_margin > list(look_up.values())[spread_next_i]:  # 如果节点i的Inf Spread大于上一轮lookup表中的最前面的节点的Inf Spread，直接加入种子集
                    S.append(i)
                    flag = 1
                    spread = spread_i
                    SPREAD.append(spread)
                    timelapse.append(time.time() - start_time)
                    break
                elif spread_i > spread_max:
                    spread_max = s
                    node_max = i

                look_up[i] = spread_i_margin

        if flag == 0:
            S.append(node_max)
            spread = spread_max
            SPREAD.append(spread)
            timelapse.append(time.time() - start_time)
    return (S, spread, timelapse)

data = pd.read_csv('../data/facebook1.csv')
p = generate_p(data.shape[0])

(S, spread, timelapse) = CELF(data, k=3, mc=1)

print(S, spread, timelapse)