成都创新互联网站制作重庆分公司

文本挖掘分词mapreduce化

软件版本

让客户满意是我们工作的目标,不断超越客户的期望值来自于我们对这个行业的热爱。我们立志把好的技术通过有效、简单的方式提供给客户,将通过不懈努力成为客户在信息化领域值得信任、有价值的长期合作伙伴,公司提供的服务项目有:申请域名虚拟主机、营销软件、网站建设、襄城网站维护、网站推广。

paoding-analysis3.0

文本挖掘分词mapreduce化

项目jar包和拷贝庖丁dic目录到项目的类路径下

文本挖掘分词mapreduce化

修改paoding-analysis.jar下的paoding-dic-home.properties文件设置词典文件路径

paoding.dic.home=classpath:dic

分词程序demo

import java.io.IOException;
import java.io.StringReader;

import org.apache.lucene.analysis.TokenStream;
import org.apache.lucene.analysis.tokenattributes.CharTermAttribute;

import net.paoding.analysis.analyzer.PaodingAnalyzer;


public class TokenizeWithPaoding {
public static void main(String[] args) {
    
    String line="中华民族共和国";
    PaodingAnalyzer analyzer =new PaodingAnalyzer();
    StringReader sr=new StringReader(line);
    TokenStream ts=analyzer.tokenStream("", sr);//分词流,第一个参数无意义
    //迭代分词流
    try {
        while(ts.incrementToken()){
            CharTermAttribute ta=ts.getAttribute(CharTermAttribute.class);
            System.out.println(ta.toString());
        }
    } catch (Exception e) {
        
        e.printStackTrace();
    }
}
}

新闻文文本分类源文件

http://people.csail.mit.edu/jrennie/20Newsgroups/20news-bydate.tar.gz

每个文件夹代表一个类别,每个类别下的文件代表一条新闻

文本挖掘分词mapreduce化

中文新闻分类需要先分词

对于大量小文件可以使用FileInputFormat的另一个抽象子类CombineFileInputFormat实现createRecordReader方法

CombineFileInputFormat重写了getSpilt方法,返回的分片类型是CombineFileSpilt,是InputSpilt的子类,可包含多个文件

RecordReader怎么由文件生成key-value是由nextKeyValue函数决定

自定义的CombineFileInputFormat类

package org.conan.myhadoop.fengci;



import java.io.IOException;

import org.apache.hadoop.fs.Path;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.mapreduce.InputSplit;
import org.apache.hadoop.mapreduce.JobContext;
import org.apache.hadoop.mapreduce.RecordReader;
import org.apache.hadoop.mapreduce.TaskAttemptContext;
import org.apache.hadoop.mapreduce.lib.input.CombineFileInputFormat;
import org.apache.hadoop.mapreduce.lib.input.CombineFileRecordReader;
import org.apache.hadoop.mapreduce.lib.input.CombineFileSplit;

/**
 * 自定义MyInputFormat类, 用于实现一个Split包含多个文件
 * @author BOB
 *
 */
public class MyInputFormat extends CombineFileInputFormat{
        
        //禁止文件切分
        @Override
        protected boolean isSplitable(JobContext context, Path file) {
                return false;
        }

        @Override
        public RecordReader createRecordReader(InputSplit split, TaskAttemptContext context) throws IOException {
                return new CombineFileRecordReader((CombineFileSplit)split, context, MyRecordReader.class);
        }

}

自定义的RecordReader类

package org.conan.myhadoop.fengci;

import java.io.IOException;

import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.fs.FSDataInputStream;
import org.apache.hadoop.fs.FileSystem;
import org.apache.hadoop.fs.Path;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.mapreduce.InputSplit;
import org.apache.hadoop.mapreduce.RecordReader;
import org.apache.hadoop.mapreduce.TaskAttemptContext;
import org.apache.hadoop.mapreduce.lib.input.CombineFileSplit;

/**
 * 自定义MyRecordReader类, 用于读取MyInputFormat对象切分的Split分片中的内容
 * @author BOB
 *
 */
public class MyRecordReader extends RecordReader {

        private CombineFileSplit combineFileSplit;                //当前处理的分片
        private Configuration conf;                        //作业的配置信息
        private Text currentKey = new Text();                //当前读入的key
        private Text currentValue = new Text();        //当前读入的value
        private int totalLength;                        //当前分片中文件的数量
        private int currentIndex;                                //正在读取的文件在当前分片中的位置索引
        private float currentProgress = 0F;                //当前进度
        private boolean processed = false;        //标记当前文件是否已经被处理过
        
        //构造方法
        public MyRecordReader(CombineFileSplit combineFileSplit,
                        TaskAttemptContext context, Integer fileIndex) {
                super();
                this.combineFileSplit = combineFileSplit;
                this.currentIndex = fileIndex;
                this.conf = context.getConfiguration();
                this.totalLength = combineFileSplit.getPaths().length;
        }

     
        @Override
        public void initialize(InputSplit split, TaskAttemptContext context) throws IOException, InterruptedException {

        }
        @Override
        public Text getCurrentKey() throws IOException, InterruptedException {
                return currentKey;
        }

        @Override
        public Text getCurrentValue() throws IOException, InterruptedException {
                return currentValue;
        }

        @Override
        public float getProgress() throws IOException, InterruptedException {
                if(currentIndex >= 0 && currentIndex < totalLength) {
                        return currentProgress = (float) currentIndex/totalLength;
                }
                return currentProgress;
        }

        @Override
        public void close() throws IOException {

        }


        @Override
        public boolean nextKeyValue() throws IOException, InterruptedException {
                if(!processed) {
                        //由文件的父目录, 文件名以及目录分割符组成key
                        Path file = combineFileSplit.getPath(currentIndex);
                        StringBuilder sb = new StringBuilder();
                        sb.append("/");
                        sb.append(file.getParent().getName()).append("/");
                        sb.append(file.getName());
                        currentKey.set(sb.toString());
                        
                        //以整个文件的内容作为value
                        FSDataInputStream in = null;
                        byte[] content = new byte[(int)combineFileSplit.getLength(currentIndex)];
                        FileSystem fs = file.getFileSystem(conf);
                        in = fs.open(file);
                        in.readFully(content);
                        currentValue.set(content);
                        in.close();
                        processed = true;
                        return true;
                }
                return false;
        }

}

分词驱动类

package org.conan.myhadoop.fengci;

import java.io.IOException;
import java.io.StringReader;



import net.paoding.analysis.analyzer.PaodingAnalyzer;

import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.conf.Configured;
import org.apache.hadoop.fs.FileStatus;
import org.apache.hadoop.fs.FileSystem;
import org.apache.hadoop.fs.FileUtil;
import org.apache.hadoop.fs.Path;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.mapreduce.Job;
import org.apache.hadoop.mapreduce.Mapper;
import org.apache.hadoop.mapreduce.lib.input.FileInputFormat;
import org.apache.hadoop.mapreduce.lib.output.FileOutputFormat;
import org.apache.hadoop.mapreduce.lib.output.SequenceFileOutputFormat;
import org.apache.hadoop.util.Tool;
import org.apache.hadoop.util.ToolRunner;
import org.apache.lucene.analysis.Analyzer;
import org.apache.lucene.analysis.TokenStream;
import org.apache.lucene.analysis.tokenattributes.CharTermAttribute;



/**
 * 分词驱动器类, 用于给输入文件进行分词
 * @author BOB
 *
 */
public class TokenizerDriver extends Configured implements Tool{

        
        public static void main(String[] args) throws Exception{
                int res = ToolRunner.run(new Configuration(), new TokenizerDriver(), args);
                System.exit(res);
        }

        @Override
        public int run(String[] args) throws Exception {
                Configuration conf = new Configuration();
                //参数设置
                conf.setLong("mapreduce.input.fileinputformat.split.maxsize", 4000000);
              //作业名称
                Job job = new Job(conf,"Tokenizer");
                job.setJarByClass(TokenizerDriver.class);
                
                job.setMapperClass(Map.class);
                
                job.setInputFormatClass(MyInputFormat.class);
                
                job.setOutputFormatClass(SequenceFileOutputFormat.class);
                
                job.setOutputKeyClass(Text.class);
                job.setOutputValueClass(Text.class);
                
                Path inpath=new Path(args[0]);
                Path outpath=new Path(args[1]);
                FileSystem fs = inpath.getFileSystem(conf);
                FileStatus[] status = fs.listStatus(inpath);
                Path[] paths = FileUtil.stat2Paths(status);
                for(Path path : paths) {
                        FileInputFormat.addInputPath(job, path);
                }
                FileOutputFormat.setOutputPath(job, outpath);
                
                //输出文件夹已经存在则删除
                FileSystem hdfs = outpath.getFileSystem(conf);
                if(hdfs.exists(outpath)){
                    hdfs.delete(outpath,true);
                    hdfs.close();
                }
                //没有Reduce任务
                job.setNumReduceTasks(0); 
                return job.waitForCompletion(true) ? 0 : 1;
        }
        
        /**
         * Hadoop计算框架下的Map类, 用于并行处理文本分词任务
         * @author BOB
         *
         */
        static class Map extends Mapper {
                
                @Override
                protected void map(Text key, Text value, Context context) throws IOException, InterruptedException {
                        //创建分词器
                        Analyzer analyzer = new PaodingAnalyzer();
                        String line = value.toString();
                        StringReader reader = new StringReader(line);
                        //获取分词流对象
                        TokenStream ts = analyzer.tokenStream("", reader);
                        StringBuilder sb = new StringBuilder();
                        
                        //遍历分词流中的词语
                        while(ts.incrementToken()) {
                                CharTermAttribute ta = ts.getAttribute(CharTermAttribute.class);
                                if(sb.length() != 0) {
                                        sb.append(" ").append(ta.toString());
                                } else {
                                        sb.append(ta.toString());
                                }
                        }
                        value.set(sb.toString());
                        context.write(key, value);
                }
                
        }
}

分词预先处理结果,将所有新闻集中到一个文本中,key为类别,一行代表一篇新闻,单词之间用空格分开

文本挖掘分词mapreduce化

处理后的数据可用于mahout做贝叶斯分类器

参考文章:

http://f.dataguru.cn/thread-244375-1-1.html

http://www.cnblogs.com/panweishadow/p/4320720.html


 


当前名称:文本挖掘分词mapreduce化
分享网址:http://cxhlcq.com/article/jsedjj.html

其他资讯

在线咨询

微信咨询

电话咨询

028-86922220(工作日)

18980820575(7×24)

提交需求

返回顶部