C++实现PageRank（标准、增量）

标准PageRank

原理

$$PR(v_j^k)=d\ast \sum _{v_i\in in(v_j)}\frac{PR(v_i^{k-1})}{|OUT(v_i)|}+(1-d)$$

代码

#include <iostream>
#include <fstream>
#include <vector>
#include <set>
using namespace std;

double d = 0.85;
double threshold = 1e-7;

struct Page
{
    vector<int> outPage; //出邻居
    double oldPR = 0;
    double newPR = 0;
};

//读文件，获取顶点个数
int getN()
{
    set<int> s;
    ifstream inFile("/Users/joy/git-repo/PageRank-cpp/a.txt");
    int u, v;
    while(inFile >> u >> v) //假设每一行插入一条边u->v
    {
        s.insert(u);
        s.insert(v);
    }
    int N = s.size();
    return N;
}

//把每个页面的信息（outPage，oldPR，newPR）存到向量里
void getPagesVec(int N, vector<Page>& pages)
{
    ifstream inFile("/Users/joy/git-repo/PageRank-cpp/a.txt");
    int u, v;
    while(inFile >> u >> v)
        pages[u].outPage.push_back(v);
}

int main()
{
    int N = getN(); // N是页面总数
    vector<Page> pages(N); //pages存放页面向量，大小为N
    getPagesVec(N, pages);

    int cnt = 0; //统计迭代几轮
	int shouldStop = 0; //根据oldPR与newPR的差值 判断是否停止迭代
    // 开始pr值的迭代更新
    while(!shouldStop)
    {
        shouldStop = 1;

        //对于i->j，根据公式，页面j能接收到的PR值为：PR(j) = d * sum(PR(i)/OUT(i)) + (1-d)
        //因此每个顶点i向出页面发送的PR值为：oldPR/outDegree
        for(auto &page: pages)
        {
            int outDegree = page.outPage.size();
            double tmpPR = page.oldPR / outDegree;
            for(int j = 0; j < outDegree; j++)
                pages[page.outPage[j]].newPR += tmpPR * d;
        }

        for(auto &page: pages)
        {
            page.newPR += 1-d;
            if(abs(page.newPR - page.oldPR) < threshold)
                shouldStop = 0;
            page.oldPR = page.newPR;
            page.newPR = 0;
        }

        cnt++;
    }

    printf("%s%d%s\n", "共迭代", cnt, "轮");
    for(auto page: pages)
        printf("%.9lf\n", page.oldPR/N);

    return 0;
}

使用的数据集

0 2
0 1
0 3
1 0
1 3
2 0
3 1
3 2

运行结果

共迭代90轮
0.324561278
0.225146115
0.225146115
0.225146115

增量PageRank

原理

$$\begin{gathered} delta(v_j^k)=\sum _{v_i\in in(v_j)}\frac{delta(v_i^{k-1})}{|OUT(v_i)|}\ast d \\ value(v_j^k)+=delta(v_j^k) \end{gathered}$$

代码

#include <iostream>
#include <fstream>
#include <vector>
#include <set>
using namespace std;

double d = 0.85;
double threshold = 1e-7;

struct Page
{
    vector<int> outPage; //出邻居
    double value = 0;
    double oldDelta= 0;
    double recvDelta = 1-d;
};

//读文件，获取顶点个数
int getN()
{
    set<int> s;
    ifstream inFile("/Users/joy/git-repo/PageRank-cpp/a.txt");
    int u, v;
    while(inFile >> u >> v) //假设每一行插入一条边u->v
    {
        s.insert(u);
        s.insert(v);
    }
    int N = s.size();
    return N;
}

//把每个页面的信息（outPage，value，oldDelta，recvDelta）存到向量里
void getPagesVec(int N, vector<Page>& pages)
{
    ifstream inFile("/Users/joy/git-repo/PageRank-cpp/a.txt");
    int u, v;
    while(inFile >> u >> v)
        pages[u].outPage.push_back(v);
}

int main()
{
    int N = getN(); // N是页面总数
    vector<Page> pages(N); //pages存放页面向量，大小为N
    getPagesVec(N, pages);

    int cnt = 0; //统计迭代几轮
	int shouldStop = 0; //根据oldPR与newPR的差值 判断是否停止迭代
    // 开始迭代
    while(!shouldStop)
    {
        shouldStop = 1;

        //对于i->j，根据公式，页面j能接收到的delta值为：
        //recvDelta(j) = oldDelta(i) / outDegree
        //因此每个顶点i向出邻居发送的delta值为：oldDelta(i)
        for(auto &page: pages)
        {
            int outDegree = page.outPage.size();
            double tmpDelta = page.oldDelta / outDegree;
            for(int j = 0; j < outDegree; j++)
                pages[page.outPage[j]].recvDelta += tmpDelta * d;
        }

        for(auto &page: pages)
        {
            if(page.recvDelta < threshold)
                shouldStop = 0;
            page.value += page.recvDelta;
            page.oldDelta = page.recvDelta;
            page.recvDelta = 0;
        }
        
        cnt++;
    }

    printf("%s%d%s\n", "共迭代", cnt, "轮");
    for(auto page: pages)
        printf("%.9lf\n", page.value/N);

    return 0;
}

使用的数据集

0 2
0 1
0 3
1 0
1 3
2 0
3 1
3 2

运行结果

共迭代90轮
0.324561278
0.225146115
0.225146115
0.225146115

使用networkx进行验证

代码

以下这个利用python的第三方库networkx实现的pagerank算法可以用来测试自己代码正确性：

import networkx as nx

filename = 'dataset/a.txt'
G = nx.DiGraph()
with open(filename) as file:
    for line in file:
        head, tail = [int(x) for x in line.split()]
        G.add_edge(head, tail)

pr = nx.pagerank(G)
print(pr)

运行结果

{0: 0.3245609358176832, 
2: 0.22514635472743894, 
1: 0.22514635472743894, 
3: 0.22514635472743896}