Naive Bayesian文本分类器
jopen
9年前
贝叶斯学习方法中实用性很高的一种为朴素贝叶斯学习期，常被称为朴素贝叶斯分类器。在某些领域中与神经网络和决策树学习相当。虽然朴素贝叶斯分类器忽略单词间的依赖关系，即假设所有单词是条件独立的，但朴素贝叶斯分类在实际应用中有很出色的表现。
朴素贝叶斯文本分类算法伪代码：
朴素贝叶斯文本分类算法流程：
通过计算训练集中每个类别的概率与不同类别下每个单词的概率，然后利用朴素贝叶斯公式计算新文档被分类为各个类别的概率，最终输出概率最大的类别。
C++源码：
/* Bayesian classifier for document classifiaction 15S103182 Ethan 2015.12.27 */ #include <iostream> #include <vector> #include <iterator> #include <map> #include <fstream> #include <iomanip> #include <sstream> using namespace std; int stringToInteger(string a){ stringstream ss; ss<<a; int b; ss>>b; return b; } vector<int> openClassificationFile(const char* dataset){ fstream file; file.open(dataset,ios::in); if(!file) { cout <<"Open File Failed!" <<endl; vector<int> a; return a; } vector<int> data; int i=1; while(!file.eof()){ string temp; file>>temp; data.push_back(stringToInteger(temp)); } file.close(); return data; } vector<string> openFile(const char* dataset){ fstream file; file.open(dataset,ios::in); if(!file) { cout <<"Open File Failed!" <<endl; vector<string> a; return a; } vector<string> data; int i=1; while(!file.eof()){ string temp; file>>temp; data.push_back(temp); } file.close(); for(int i=0;i<data.size();i++) cout<<data[i]<<"\t"; cout<<endl; cout<<"Open file successfully!"<<endl; return data; } vector<vector<string> > openFiles(const vector<char*> files){ vector<vector<string> > docs; for(int i=0;i<files.size();i++){ vector<string> t = openFile(files[i]); docs.push_back(t); } return docs; } void bayesian(vector<vector<string> > docs,vector<int> c,vector<string> d){ map<string,int> wordFrequency;//每个单词出现的个数 map<int,float> cWordProbability;//类别单词频率 map<int,int> cTotalFrequency;//类别单词个数 map<int,map<string,int> > cWordlTotalFrequency;//类别下单词个数 int totalWords=0; for(int i=0;i<docs.size();i++){ totalWords += docs[i].size(); cWordProbability[c[i]] = cWordProbability[c[i]] + docs[i].size(); map<string,int> sn; for(int j=0;j<docs[i].size();j++){ wordFrequency[docs[i][j]] = wordFrequency[docs[i][j]] + 1; sn[docs[i][j]] = sn[docs[i][j]] + 1; } map<string,int>::iterator isn; for(isn = sn.begin();isn!=sn.end();isn++){ cWordlTotalFrequency[c[i]][isn->first] = cWordlTotalFrequency[c[i]][isn->first] + isn->second; } } int tw = wordFrequency.size(); map<int,float>::iterator icWordProbability; for(icWordProbability=cWordProbability.begin();icWordProbability!=cWordProbability.end();icWordProbability++){ cTotalFrequency[icWordProbability->first] = icWordProbability->second; cWordProbability[icWordProbability->first] = icWordProbability->second / totalWords; } cout<<"Word Frequency:"<<endl; map<string,int>::iterator iwordFrequency; for(iwordFrequency=wordFrequency.begin();iwordFrequency!=wordFrequency.end();iwordFrequency++){ cout<<setw(8)<<iwordFrequency->first<<"\tFrequency:"<<iwordFrequency->second<<endl; } cout<<"Conditional Probability:"<<endl; map<string,int> dtw;//待分类文档词频 for(int i=0;i<d.size();i++) dtw[d[i]] = dtw[d[i]] + 1; map<string,map<int,float> > cp;//单词类别概率 map<string,int>::iterator idtw; for(idtw=dtw.begin();idtw!=dtw.end();idtw++){ map<int,float> cf; for(int j=0;j<cTotalFrequency.size();j++){ float p=0; p = (float)(cWordlTotalFrequency[j][idtw->first] +1) / (cTotalFrequency[j] + wordFrequency.size()); cf[j] = p; cout<<"P("<<idtw->first<<"|"<<j<<") \t= "<<p<<endl; } cp[idtw->first] = cf; } cout<<"Classification Probability:"<<endl; float mp = 0; int classification=0; for(int i=0;i<cTotalFrequency.size();i++){ float tcp=1; for(int j=0;j<d.size();j++){ tcp = tcp * cp[d[j]][i]; } tcp = tcp * cWordProbability[i]; cout<<"classification:"<<i<<"\t"<<"Probability:"<<tcp<<endl; if(mp<tcp) { mp = tcp; classification = i; } } cout<<"The new document classification is："<<classification<<endl; } int main(int argc, char** argv) { vector<vector<string> > docs; vector<int> c = openClassificationFile("classification.txt"); vector<char *> files; files.push_back("1.txt");files.push_back("2.txt");files.push_back("3.txt");files.push_back("4.txt");files.push_back("5.txt"); cout<<"训练文档集："<<endl; docs = openFiles(files); vector<string> d; cout<<"待分类文档："<<endl; d = openFile("new.txt"); bayesian(docs,c,d); return 0; }效果展示：
结论：
朴素贝叶斯分类器用于处理离散型的文本数据，能够有效对文本文档进行分类。在实验过程中，最困难的地方在于数据结构的设计，由于要统计每个文档类别的频数和每个文档类别下单词的概率，这个地方需要用到复杂映射与统计，在编码过程中经过不断的思考，最终通过多级映射的形式储存所需的数据，最终计算出新文档的类别。通过实验，成功将新的未分类文档输入例子分类为期待的文档类型，实验结果较为满意。