简介

我们首先应该从WAP-Tree说起,下面一段话摘自《Effective Web Log Mining using WAP Tree-Mine》原文

Abstract -World Wide Web is a huge data repository and is growing with the explosive rate of about 1 million pages a day,web log records each access of the web
page and number of entries in the web logs is increasing rapidly. These web logs,when mined properly can provide useful information for decision-making. Sequential pattern mining discovers frequent user access patterns from web logs. Since Apriori-like sequential
pattern mining techniques requires expensive multiple scans of database. But, recently a novel data structure, known as Web Access Pattern Tree (or WAP-tree), was developed. This proposed method an efficient WAP-tree mining algorithm,known as DLT-mine (Doubly
Linked Tree algorithm). Proposed recursive algorithm uses this doubly Linked tree to efficiently find all access patterns that satisfy user specified criteria. This mining algorithm is faster than the other Apriori-based mining algorithms.
这段话的大致意思就是:互联网数据十分巨大,如今每天大约有1百万的网页访问增加量,并且增长十分迅速。而这些访问的信息可能给我们提供许多有用的信息并且用来制定相应的决策,并且传统的搜索信息的方法并不是十分高效,于是后来一个神奇的数据结构,称为“Web Access Pattern Tree”(简称WAP-Tree)被人们提出来了。显然,它具有更优秀的性质,并比以往的方法要更快。

后来在实现WAP-Tree的算法的过程中,人们发现WAP-Tree在搜索频繁项的过程中还可以更进一步的优化,于是人们将它改进后成为“Pre-Order Linked WAP-Tree”(简称PLWAP-Tree),具体内容我们会在下面陈述。

 
WAP-Tree
首先给出伪代码
树的构建:
A. Algorithm 2 (Doubly Linked Tree Construction)
Input: A Web access sequence database WAS and a set of all possible events E.
Output: A doubly linked tree T.
Method:
Scan 1:
1. For each access sequence S of the WAS
1.1. For each event in E
1.1.1. For each event of an access sequence of WAS. If selected event of access sequence is equal to selected event of E then
a. event count = event count + 1
b. continue with the next event in E.
2. For each event in E if event qualify the threshold add that event in the set of frequent event FE. Scan 2:
1. Create a root node for T
2. For each access sequence S in the access sequence database WAS do
(a) Extract frequent subsequence S’ from S by removing all events appearing in S but not in FE. Let S' = s1s2….sn , where si (1≤ i ≤ n) are events in S’. Let current node is a pointer that is currently pointing to the root of T.
(b) For i=1 to n do, if current node has a child labeled si , increase the count of si by 1 and make current node point to si , else create a new child node with label= si , count =1, parent pointer = current node and make current node point to the new node, and insert it into the si -queue
3. Return (T);
其中Scan1首先计算输入序列的每个字符出现的频度,Scan2筛除其中频度低于阀值lamda的字符,然后构建一颗字典树T,同时要让上一个Si指向此Si,最后返回这颗树。
最后这颗树看起来应该是这个样子的
TID Web access sequence Frequent subsequence
100 abdac abac
200 eaebcac abcac
300 babfaec babac
400 afbacfc abacc

最后在树中进行搜索频发序列,伪代码如下:

B. Algorithm 2 (Mining all ξ-patterns in doubly linked tree)
Input: a Doubly linked tree T and support threshold ξ.
Output: the complete set of ξ-patterns.
Method:
1. If doubly linked tree T has only one branch, return all the unique combinations of nodes in that branch
2. Initialize Web access pattern set WAP=φ. Every event in T itself is a Web access pattern, insert them into WAP
3. For each event ei in T,
a. Construct a conditional sequence base of ei , i.e.PS( ei ), by following the ei -queue, count conditional frequent events at the same time.
b. If the set of conditional frequent events is not empty, build a conditional doubly linked tree for ei over PS( ei ) using algorithm 1. Recursively mine the conditional doubly linked tree
c. For each Web access pattern returned from mining the conditional doubly linked tree, concatenate ei to it and insert it into WAP.
4. Return WAP.

最后我们就会得到频繁项如下:

{c, aac, bac, abac, ac, abc, bc, b, ab, a, aa,ba, aba}

PLWAP-Tree

人们在运用WAP-Tree的过程中,发现其在时间复杂度上并不理想,请看原文《PLWAP Sequential Mining: Open Source Code》中对PLWAP-Tree的一段介绍:

Abstract -PLWAP algorithm uses a preorder linked, position code dversion of WAP tree and eliminates the need to recursively re-construct intermediate WAP trees
during sequential mining as done by WAP tree technique. PLWAP produces sig-nificant reduction in response time achieved by the WAP algorithm and provides a position code mechanism for remembering the stored database, thus, eliminating the need to re-scan the
original database as would be necessary for applications like those incrementally maintaining mined frequent patterns, performing stream or dynamic mining.
大致意思就是:PLWAP 算法使用先序遍历整个树来建立Head-Table链表队列,并且为每个节点设置一个独一无二的编号,并且可以根据这个编号立刻知道一个节点是不是另一个节点的子节点。
最后相同数据下PLWAP-Tree构造如图:

看图应该很容易懂,这里提示几点方便大家理解:

1、上图中比如{c:1:1110}表示这个节点代表的字符是c,而其权重是1,即只有1个c,而1110表示这个节点的编号。编号规则是

①根节点编号为空

②对于节点u其编号为s,设其子节点从左到右分别为v1,v2,v3……,则其编号分别s1,s10,s100……以此类推,即每次多一个0

这样判断p是否是q的后辈点的方法就是:在q的后面加一个“1”,然后判断是否是p的前缀,如果是则p是q的后辈节点

2、关于Head-Table,在PLWAP-Tree中其是在整棵树构建成功后再构建PLWAP-Tree链表的(和WAP-Tree的不同,希望大家好好体会),构建的方案是按照先序遍历的顺序(上图的虚线部分)。大家可以和WAP-Tree的Head-Table的虚线箭头做一下对比,很容易就能发现它们的区别。

PLWAP-Tree代码实现(c++)

这里放上我自己实现的PLWAP-Tree代码,供给大家参考

#include <stdio.h>
#include <tchar.h>
#include <string>
#include <cstring>
#include <vector>
#include <iostream>
#include <string>
#include <map> #define alp_maxn 130 using namespace std; struct Node{
char alp;
int alp_count;
struct Node * nex;
vector<struct Node*>son;
string seq;
Node(int _siz, char _alp);
}; class PLWAPTREE{
private:
Node * root; //the root of the plwap-tree
Node * Head_Table[alp_maxn]; //Head_Table
Node * alp_las[alp_maxn];
int lamda; //lamda int alp_tot; //the number of valid words
char alp_link[alp_maxn]; //discratization
int alp_count[alp_maxn]; //discratization
map<char, int>alp_translate; //discratization public: vector<string>reads;
vector<string>feq; //the frequent words void Init(int _lamda);
void AddString(string st);
void BuildTree();
void BuildTree(Node *s, string id);
void SearchFeq(vector<string>R, string now_feq); void print_tree(Node *s); //debug only...
Node * get_root(); //debug only...
}; Node * PLWAPTREE::get_root(){
return root;
} void PLWAPTREE::print_tree(Node *s){
if (s == NULL) return;
cout << "char : " << s->alp << " seq : " << s->seq << " alp_count : " << s->alp_count;
if (s->nex != NULL) cout << " nex_seq :" << s->nex->seq << endl;
else cout << endl;
for (int i = 0; i < alp_tot; i++)
print_tree(s->son[i]);
} Node::Node(int _siz, char _alp = -1){
nex = NULL;
son.clear();
while (_siz--) {
son.push_back(NULL);
}
alp = _alp;
alp_count = 0;
} void PLWAPTREE::Init(int _lamda){
root = new Node(alp_maxn);
for (int i = 0; i < alp_maxn; i++){
Head_Table[i] = NULL;
alp_count[i] = 0;
alp_las[i] = NULL;
}
reads.clear();
feq.clear();
alp_translate.clear();
alp_tot = 0;
lamda = _lamda;
} void PLWAPTREE::AddString(string st){
int alp_tmp[alp_maxn];
memset(alp_tmp, 0, sizeof(alp_tmp));
for (int i = 0; i < st.length(); i++)
alp_tmp[(int)st[i]] = 1;
for (int i = 0; i < alp_maxn; i++)
alp_count[i] += alp_tmp[i];
reads.push_back(st);
} void PLWAPTREE::BuildTree(){
for (int i = 0; i < alp_maxn; i++){
if (alp_count[i] >= lamda){
alp_link[alp_tot] = (char)i;
alp_translate[(char)i] = alp_tot;
alp_tot++;
}
} //discretization to save memory and time printf("-discretization success !\n"); for (int i = 0; i < reads.size(); i++){
string now_string = reads[i];
Node * pnow = root;
for (int j = 0; j < now_string.length(); j++){
if (alp_count[(int)now_string[j]] < lamda) continue;
int sig = alp_translate[now_string[j]];
if (pnow->son[sig] == NULL){
Node * tmp = new Node(alp_tot, now_string[j]);
pnow->son[sig] = tmp;
}
pnow = pnow->son[sig];
pnow->alp_count++;
}
} printf("-trip-build success !\n"); BuildTree(root, "");
} void PLWAPTREE::BuildTree(Node *s, string id){
string seq = id + "1";
for (int i = 0; i < alp_tot; i++){
if (s->son[i] == NULL) continue;
if (Head_Table[i] == NULL){
Head_Table[i] = s->son[i];
}
if (alp_las[i] != NULL){
alp_las[i]->nex = s->son[i];
}
alp_las[i] = s->son[i];
s->son[i]->seq = seq;
BuildTree(s->son[i], seq);
seq = seq + "0";
}
} void PLWAPTREE::SearchFeq(vector<string>R, string now_feq){
for (int i = 0; i < alp_tot; i++){
Node * p = Head_Table[i];
bool flag = true;
if (R.size() != 0){
flag = false;
while (p != NULL){
for (int j = 0; j < R.size(); j++){
string str = R[j] + "1";
int sig = p->seq.find(str);
if (sig == 0){
flag = true;
break;
}
}
if (flag) break;
p = p->nex;
}
}
if (flag == false) continue; int C = p->alp_count;
string S = p->seq;
vector<string>Rs; Rs.clear();
Rs.push_back(p->seq); for (p = p->nex; p != NULL; p = p->nex){
bool is_son_of_R = false;
bool is_son_of_S = false;
if (R.size() == 0) is_son_of_R = true;
else{
for (int j = 0; j < R.size(); j++){
string str = R[j] + "1";
int sig = p->seq.find(str);
if (sig == 0){
is_son_of_R = true;
break;
}
}
}
string str = S + "1";
int sig = p->seq.find(str);
if (sig == 0){
is_son_of_S = true;
}
if (is_son_of_R == true && is_son_of_S == false){
C += p->alp_count;
Rs.push_back(p->seq);
S = p->seq;
}
} if (C >= lamda){
feq.push_back(now_feq + alp_link[i]);
SearchFeq(Rs, now_feq + alp_link[i]);
}
}
} int main(){
PLWAPTREE pt;
pt.Init(3); printf("Init success !\n"); pt.AddString("abdac");
pt.AddString("eaebcac");
pt.AddString("babfaec");
pt.AddString("afbacfc"); printf("read string success !\n"); pt.BuildTree(); printf("Buile tree success !\n");
/*
printf("tree just like :\n");
pt.print_tree(pt.get_root());
*/ vector<string>tmp; tmp.clear();
pt.SearchFeq(tmp, ""); printf("result : \n"); for (int i = 0; i < pt.feq.size(); i++)
cout << pt.feq[i] << endl; getchar();
return 0;
}
参考资料:
《Effective Web Log Mining using WAP Tree-Mine》
《PLWAP Sequential Mining: Open Source Code ∗》

数据挖掘:WAP-Tree与PLWAP-Tree的更多相关文章

  1. 【数据挖掘】分类之decision tree(转载)

    [数据挖掘]分类之decision tree. 1. ID3 算法 ID3 算法是一种典型的决策树(decision tree)算法,C4.5, CART都是在其基础上发展而来.决策树的叶子节点表示类 ...

  2. B-Tree、B+Tree和B*Tree

    B-Tree(这儿可不是减号,就是常规意义的BTree) 是一种多路搜索树: 1.定义任意非叶子结点最多只有M个儿子:且M>2: 2.根结点的儿子数为[2, M]: 3.除根结点以外的非叶子结点 ...

  3. 【Luogu1501】Tree(Link-Cut Tree)

    [Luogu1501]Tree(Link-Cut Tree) 题面 洛谷 题解 \(LCT\)版子题 看到了顺手敲一下而已 注意一下,别乘爆了 #include<iostream> #in ...

  4. 【BZOJ3282】Tree (Link-Cut Tree)

    [BZOJ3282]Tree (Link-Cut Tree) 题面 BZOJ权限题呀,良心luogu上有 题解 Link-Cut Tree班子提 最近因为NOIP考炸了 学科也炸了 时间显然没有 以后 ...

  5. [LeetCode] Encode N-ary Tree to Binary Tree 将N叉树编码为二叉树

    Design an algorithm to encode an N-ary tree into a binary tree and decode the binary tree to get the ...

  6. 平衡二叉树(Balanced Binary Tree 或 Height-Balanced Tree)又称AVL树

    平衡二叉树(Balanced Binary Tree 或 Height-Balanced Tree)又称AVL树 (a)和(b)都是排序二叉树,但是查找(b)的93节点就需要查找6次,查找(a)的93 ...

  7. WPF中的Visual Tree和Logical Tree与路由事件

    1.Visual Tree和Logical TreeLogical Tree:逻辑树,WPF中用户界面有一个对象树构建而成,这棵树叫做逻辑树,元素的声明分层结构形成了所谓的逻辑树!!Visual Tr ...

  8. 笔试算法题(39):Trie树(Trie Tree or Prefix Tree)

    议题:TRIE树 (Trie Tree or Prefix Tree): 分析: 又称字典树或者前缀树,一种用于快速检索的多叉树结构:英文字母的Trie树为26叉树,数字的Trie树为10叉树:All ...

  9. LC 431. Encode N-ary Tree to Binary Tree 【lock,hard】

    Design an algorithm to encode an N-ary tree into a binary tree and decode the binary tree to get the ...

  10. 将百分制转换为5分制的算法 Binary Search Tree ordered binary tree sorted binary tree Huffman Tree

    1.二叉搜索树:去一个陌生的城市问路到目的地: for each node, all elements in its left subtree are less-or-equal to the nod ...

随机推荐

  1. 移动端调试 — 安卓机+chrome

    移动端开发时,我们常使用chrome自带的模拟器,模拟各种手机设备. 但模拟毕竟是模拟,当开发完毕,使用真机访问页面出现问题时如何调试呢? 下面介绍一种针对Android机的调试方法 1. 在pc和a ...

  2. 130、TensorFlow操作多个计算图

    # Programming with multiple graphs # 当训练一个模型的时候一个常用的方式就是使用一个图来训练你的模型 # 另一个图来评价和计算训练的效果 # 在许多情况下前向计算和 ...

  3. Discuz升级 Database Error : pre_common_syscache ADD PRIMARY KEY (cname)【解决办法】

    错误码: 1068Multiple primary key defined Execution Time : 00:00:00:000Transfer Time : 00:00:00:000Total ...

  4. stl应用(map)或字典树(有点东西)

    M - Violet Snow Gym - 101350M Every year, an elephant qualifies to the Arab Collegiate Programming C ...

  5. Linux之systemd服务配置及自动重启

    layout: post title: Linux之systemd服务配置及自动重启 date: 2019-09-09 tags: linux --- Linux之systemd服务配置及自动重启 0 ...

  6. rabbitmq部署安装

    1.安装rabbitmq需要提前安装erlang,相关下载地址:http://www.erlang.org/downloads,需要注意:默认安装的RabbitMQ 监听端口是5672     启动和 ...

  7. ORM多对多的实现

    #coding=utf-8 from sqlalchemy import Table, Column, Integer,String,DATE, ForeignKey from sqlalchemy. ...

  8. 并行开发 1.Parallel

    原文:8天玩转并行开发——第一天 Parallel的使用 随着多核时代的到来,并行开发越来越展示出它的强大威力,像我们这样的码农再也不用过多的关注底层线程的实现和手工控制, 要了解并行开发,需要先了解 ...

  9. k3 cloud查看附件提示授予目录NetWorkService读写权限

    打开文件的时候出现下面的提示: 解决办法: 解决办法:找到C:\Program Files(x86)\Kingdee\K3Cloud\WebSite\FileUpLoadServices,在下面创建F ...

  10. js利用递归与promise 按顺序请求数据

    问题:项目中有一个需求,一个tabBar下面如果没有内容就不让该tabBar显示,当然至于有没有内容,需要我们通过请求的来判断,但是由于请求是异步的,如何让请求按照tabBar的顺序进行? 方案:我们 ...