Hadoop MapReduce 协同过滤推荐算法实战

第一步：数据预处理与用户向量构建

这一步主要清洗原始日志，将 UserID, ItemID, Pref 转换为以 User 为 Key 的向量格式。我们使用 Combiner 来减少 Shuffle 开销。

package recommand;

import java.io.IOException;
import java.util.Iterator;
import java.util.Map;

import org.apache.hadoop.fs.Path;
import org.apache.hadoop.io.IntWritable;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.mapred.FileInputFormat;
import org.apache.hadoop.mapred.FileOutputFormat;
import org.apache.hadoop.mapred.JobClient;
import org.apache.hadoop.mapred.JobConf;
import org.apache.hadoop.mapred.MapReduceBase;
import org.apache.hadoop.mapred.Mapper;
import org.apache.hadoop.mapred.OutputCollector;
import org.apache.hadoop.mapred.Reducer;
import org.apache.hadoop.mapred.Reporter;
import org.apache.hadoop.mapred.RunningJob;
import org.apache.hadoop.mapred.TextInputFormat;
import org.apache.hadoop.mapred.TextOutputFormat;

public class Step1 {

    public static class Step1_ToItemPreMapper extends MapReduceBase implements Mapper<Object, Text, IntWritable, Text> {
        private final static IntWritable k = new IntWritable();
        private final static Text v = new Text();

        @Override
        public void map(Object key, Text value, OutputCollector<IntWritable, Text> output, Reporter reporter) throws IOException {
            String[] tokens = Recommend.DELIMITER.split(value.toString());
            int userID = Integer.parseInt(tokens[0]);
            String itemID = tokens[1];
            String pref = tokens[2];
            k.set(userID);
            v.set(itemID + ":" + pref);
            output.collect(k, v);
        }
    }

    public static class Step1_ToUserVectorReducer extends MapReduceBase implements Reducer<IntWritable, Text, IntWritable, Text> {
        private final static Text v = new Text();

        @Override
        public void reduce(IntWritable key, Iterator<Text> values, OutputCollector<IntWritable, Text> output, Reporter reporter) throws IOException {
            StringBuilder sb = new StringBuilder();
            while (values.hasNext()) {
                sb.append("," + values.next());
            }
            v.set(sb.toString().replaceFirst(",", ""));
            output.collect(key, v);
        }
    }

    public static void run(Map<String, String> path) throws IOException {
        JobConf conf = Recommend.config();

        String input = path.get("Step1Input");
        String output = path.get("Step1Output");

        HdfsDAO hdfs = new HdfsDAO(Recommend.HDFS, conf);
        hdfs.rmr(output);
        hdfs.rmr(input);
        hdfs.mkdirs(input);
        hdfs.copyFile(path.get("data"), input);

        conf.setMapOutputKeyClass(IntWritable.class);
        conf.setMapOutputValueClass(Text.class);

        conf.setOutputKeyClass(IntWritable.class);
        conf.setOutputValueClass(Text.class);

        conf.setMapperClass(Step1_ToItemPreMapper.class);
        conf.setCombinerClass(Step1_ToUserVectorReducer.class);
        conf.setReducerClass(Step1_ToUserVectorReducer.class);

        conf.setInputFormat(TextInputFormat.class);
        conf.setOutputFormat(TextOutputFormat.class);

        FileInputFormat.setInputPaths(conf, new Path(input));
        FileOutputFormat.setOutputPath(conf, new Path(output));

        RunningJob job = JobClient.runJob(conf);
        while (!job.isComplete()) {
            job.waitForCompletion();
        }
    }
}

第二步：物品共现矩阵计算

基于用户向量，统计每对物品在同一用户列表中的出现次数。这是 Item-CF 的核心，用于衡量物品相似度。

package recommand;

import java.io.IOException;
import java.util.Iterator;
import java.util.Map;

import org.apache.hadoop.fs.Path;
import org.apache.hadoop.io.IntWritable;
import org.apache.hadoop.io.LongWritable;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.mapred.FileInputFormat;
import org.apache.hadoop.mapred.FileOutputFormat;
import org.apache.hadoop.mapred.JobClient;
import org.apache.hadoop.mapred.JobConf;
import org.apache.hadoop.mapred.MapReduceBase;
import org.apache.hadoop.mapred.Mapper;
import org.apache.hadoop.mapred.OutputCollector;
import org.apache.hadoop.mapred.Reducer;
import org.apache.hadoop.mapred.Reporter;
import org.apache.hadoop.mapred.RunningJob;
import org.apache.hadoop.mapred.TextInputFormat;
import org.apache.hadoop.mapred.TextOutputFormat;

public class Step2 {
    public static class Step2_UserVectorToCooccurrenceMapper extends MapReduceBase implements Mapper<LongWritable, Text, Text, IntWritable> {
        private final static Text k = new Text();
        private final static IntWritable v = new IntWritable(1);

        @Override
        public void map(LongWritable key, Text values, OutputCollector<Text, IntWritable> output, Reporter reporter) throws IOException {
            String[] tokens = Recommend.DELIMITER.split(values.toString());
            for (int i = 1; i < tokens.length; i++) {
                String itemID = tokens[i].split(":")[0];
                for (int j = 1; j < tokens.length; j++) {
                    String itemID2 = tokens[j].split(":")[0];
                    k.set(itemID + ":" + itemID2);
                    output.collect(k, v);
                }
            }
        }
    }

    public static class Step2_UserVectorToConoccurrenceReducer extends MapReduceBase implements Reducer<Text, IntWritable, Text, IntWritable> {
        IntWritable result = new IntWritable();

        @Override
        public void reduce(Text key, Iterator<IntWritable> values, OutputCollector<Text, IntWritable> output, Reporter reporter) throws IOException {
            int sum = 0;
            while (values.hasNext()) {
                sum += values.next().get();
            }
            result.set(sum);
            output.collect(key, result);
        }
    }

    public static void run(Map<String, String> path) throws IOException {
        JobConf conf = Recommend.config();

        String input = path.get("Step2Input"); 
        String output = path.get("Step2Output");

        HdfsDAO hdfs = new HdfsDAO(Recommend.HDFS, conf);
        hdfs.rmr(output);

        conf.setOutputKeyClass(Text.class);
        conf.setOutputValueClass(IntWritable.class);

        conf.setMapperClass(Step2_UserVectorToCooccurrenceMapper.class);
        conf.setCombinerClass(Step2_UserVectorToConoccurrenceReducer.class);
        conf.setReducerClass(Step2_UserVectorToConoccurrenceReducer.class);

        conf.setInputFormat(TextInputFormat.class);
        conf.setOutputFormat(TextOutputFormat.class);

        FileInputFormat.setInputPaths(conf, new Path(input));
        FileOutputFormat.setOutputPath(conf, new Path(output));

        RunningJob job = JobClient.runJob(conf);
        while (!job.isComplete()) {
            job.waitForCompletion();
        }
    }
}

第三步：数据分片准备

为了后续进行矩阵乘法，我们需要将用户评分数据和物品共现数据分别整理成适合 Join 的格式。这里分为两个并行的子任务。

package recommand;

import java.io.IOException;
import java.util.Map;

import org.apache.hadoop.fs.Path;
import org.apache.hadoop.io.IntWritable;
import org.apache.hadoop.io.LongWritable;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.mapred.FileInputFormat;
import org.apache.hadoop.mapred.FileOutputFormat;
import org.apache.hadoop.mapred.JobClient;
import org.apache.hadoop.mapred.JobConf;
import org.apache.hadoop.mapred.MapReduceBase;
import org.apache.hadoop.mapred.Mapper;
import org.apache.hadoop.mapred.OutputCollector;
import org.apache.hadoop.mapred.Reporter;
import org.apache.hadoop.mapred.RunningJob;
import org.apache.hadoop.mapred.TextInputFormat;
import org.apache.hadoop.mapred.TextOutputFormat;

public class Step3 {

    public static class Step31_UserVectorSplitterMapper extends MapReduceBase implements Mapper<LongWritable, Text, IntWritable, Text> {
        private final static IntWritable k = new IntWritable();
        private final static Text v = new Text();

        @Override
        public void map(LongWritable key, Text values, OutputCollector<IntWritable, Text> output, Reporter reporter) throws IOException {
            String[] tokens = Recommend.DELIMITER.split(values.toString());
            for (int i = 1; i < tokens.length; i++) {
                String[] vector = tokens[i].split(":");
                int itemID = Integer.parseInt(vector[0]);
                String pref = vector[1];

                k.set(itemID);
                v.set(tokens[0] + ":" + pref);
                output.collect(k, v);
            }
        }
    }

    public static void run1(Map<String, String> path) throws IOException {
        JobConf conf = Recommend.config();

        String input = path.get("Step3Input1");
        String output = path.get("Step3Output1");

        HdfsDAO hdfs = new HdfsDAO(Recommend.HDFS, conf);
        hdfs.rmr(output);

        conf.setOutputKeyClass(IntWritable.class);
        conf.setOutputValueClass(Text.class);

        conf.setMapperClass(Step31_UserVectorSplitterMapper.class);

        conf.setInputFormat(TextInputFormat.class);
        conf.setOutputFormat(TextOutputFormat.class);

        FileInputFormat.setInputPaths(conf, new Path(input));
        FileOutputFormat.setOutputPath(conf, new Path(output));

        RunningJob job = JobClient.runJob(conf);
        while (!job.isComplete()) {
            job.waitForCompletion();
        }
    }

    public static class Step32_CooccurrenceColumnWrapperMapper extends MapReduceBase implements Mapper<LongWritable, Text, Text, IntWritable> {
        private final static Text k = new Text();
        private final static IntWritable v = new IntWritable();

        @Override
        public void map(LongWritable key, Text values, OutputCollector<Text, IntWritable> output, Reporter reporter) throws IOException {
            String[] tokens = Recommend.DELIMITER.split(values.toString());
            k.set(tokens[0]);
            v.set(Integer.parseInt(tokens[1]));
            output.collect(k, v);
        }
    }

    public static void run2(Map<String, String> path) throws IOException {
        JobConf conf = Recommend.config();

        String input = path.get("Step3Input2");
        String output = path.get("Step3Output2");

        HdfsDAO hdfs = new HdfsDAO(Recommend.HDFS, conf);
        hdfs.rmr(output);

        conf.setOutputKeyClass(Text.class);
        conf.setOutputValueClass(IntWritable.class);

        conf.setMapperClass(Step32_CooccurrenceColumnWrapperMapper.class);

        conf.setInputFormat(TextInputFormat.class);
        conf.setOutputFormat(TextOutputFormat.class);

        FileInputFormat.setInputPaths(conf, new Path(input));
        FileOutputFormat.setOutputPath(conf, new Path(output));

        RunningJob job = JobClient.runJob(conf);
        while (!job.isComplete()) {
            job.waitForCompletion();
        }
    }
}

第四步：矩阵乘法与评分预测

这是最关键的一步。我们将物品共现矩阵（A）和用户评分矩阵（B）进行关联。通过 Map 端判断文件来源，分别标记为 A 类和 B 类数据，然后在 Reduce 端进行内积运算。

package recommand;

import java.io.IOException;
import java.util.HashMap;
import java.util.Iterator;
import java.util.Map;

import org.apache.hadoop.fs.Path;
import org.apache.hadoop.io.LongWritable;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.mapred.JobConf;
import org.apache.hadoop.mapreduce.Job;
import org.apache.hadoop.mapreduce.Mapper;
import org.apache.hadoop.mapreduce.Reducer;
import org.apache.hadoop.mapreduce.lib.input.FileInputFormat;
import org.apache.hadoop.mapreduce.lib.input.FileSplit;
import org.apache.hadoop.mapreduce.lib.input.TextInputFormat;
import org.apache.hadoop.mapreduce.lib.output.FileOutputFormat;
import org.apache.hadoop.mapreduce.lib.output.TextOutputFormat;

public class Step4_Update {

    public static class Step4_PartialMultiplyMapper extends Mapper<LongWritable, Text, Text, Text> {
        private String flag;

        @Override
        protected void setup(Context context) throws IOException, InterruptedException {
            FileSplit split = (FileSplit) context.getInputSplit();
            flag = split.getPath().getParent().getName();
        }

        @Override
        public void map(LongWritable key, Text values, Context context) throws IOException, InterruptedException {
            String[] tokens = Recommend.DELIMITER.split(values.toString());

            if (flag.equals("step3_2")) { // 同现矩阵
                String[] v1 = tokens[0].split(":");
                String itemID1 = v1[0];
                String itemID2 = v1[1];
                String num = tokens[1];

                Text k = new Text(itemID1);
                Text v = new Text("A:" + itemID2 + "," + num);
                context.write(k, v);
            } else if (flag.equals("step3_1")) { // 评分矩阵
                String[] v2 = tokens[1].split(":");
                String itemID = tokens[0];
                String userID = v2[0];
                String pref = v2[1];

                Text k = new Text(itemID);
                Text v = new Text("B:" + userID + "," + pref);
                context.write(k, v);
            }
        }
    }

    public static class Step4_AggregateReducer extends Reducer<Text, Text, Text, Text> {
        @Override
        public void reduce(Text key, Iterable<Text> values, Context context) throws IOException, InterruptedException {
            Map<String, String> mapA = new HashMap<>();
            Map<String, String> mapB = new HashMap<>();

            for (Text line : values) {
                String val = line.toString();
                if (val.startsWith("A:")) {
                    String[] kv = Recommend.DELIMITER.split(val.substring(2));
                    mapA.put(kv[0], kv[1]);
                } else if (val.startsWith("B:")) {
                    String[] kv = Recommend.DELIMITER.split(val.substring(2));
                    mapB.put(kv[0], kv[1]);
                }
            }

            double result = 0;
            Iterator<String> iter = mapA.keySet().iterator();
            while (iter.hasNext()) {
                String mapk = iter.next();
                int num = Integer.parseInt(mapA.get(mapk));
                Iterator<String> iterb = mapB.keySet().iterator();
                while (iterb.hasNext()) {
                    String mapkb = iterb.next();
                    double pref = Double.parseDouble(mapB.get(mapkb));
                    result = num * pref;

                    Text k = new Text(mapkb);
                    Text v = new Text(mapk + "," + result);
                    context.write(k, v);
                }
            }
        }
    }

    public static void run(Map<String, String> path) throws IOException, InterruptedException, ClassNotFoundException {
        JobConf conf = Recommend.config();

        String input1 = path.get("Step5Input1");
        String input2 = path.get("Step5Input2");
        String output = path.get("Step5Output");

        HdfsDAO hdfs = new HdfsDAO(Recommend.HDFS, conf);
        hdfs.rmr(output);

        Job job = new Job(conf);
        job.setJarByClass(Step4_Update.class);

        job.setOutputKeyClass(Text.class);
        job.setOutputValueClass(Text.class);

        job.setMapperClass(Step4_Update.Step4_PartialMultiplyMapper.class);
        job.setReducerClass(Step4_Update.Step4_AggregateReducer.class);

        job.setInputFormatClass(TextInputFormat.class);
        job.setOutputFormatClass(TextOutputFormat.class);

        FileInputFormat.setInputPaths(job, new Path(input1), new Path(input2));
        FileOutputFormat.setOutputPath(job, new Path(output));

        job.waitForCompletion(true);
    }
}

第五步：最终聚合与排序

最后一步将同一用户的不同物品得分进行累加，得到最终的推荐分数列表。

package recommand;

import java.io.IOException;
import java.util.HashMap;
import java.util.Iterator;
import java.util.Map;

import org.apache.hadoop.fs.Path;
import org.apache.hadoop.io.LongWritable;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.mapred.JobConf;
import org.apache.hadoop.mapreduce.Job;
import org.apache.hadoop.mapreduce.Mapper;
import org.apache.hadoop.mapreduce.Reducer;
import org.apache.hadoop.mapreduce.lib.input.FileInputFormat;
import org.apache.hadoop.mapreduce.lib.input.TextInputFormat;
import org.apache.hadoop.mapreduce.lib.output.FileOutputFormat;
import org.apache.hadoop.mapreduce.lib.output.TextOutputFormat;

public class Step4_Update2 {

    public static class Step4_RecommendMapper extends Mapper<LongWritable, Text, Text, Text> {
        @Override
        public void map(LongWritable key, Text values, Context context) throws IOException, InterruptedException {
            String[] tokens = Recommend.DELIMITER.split(values.toString());
            Text k = new Text(tokens[0]);
            Text v = new Text(tokens[1]+","+tokens[2]);
            context.write(k, v);
        }
    }

    public static class Step4_RecommendReducer extends Reducer<Text, Text, Text, Text> {
        @Override
        public void reduce(Text key, Iterable<Text> values, Context context) throws IOException, InterruptedException {
            Map<String, Double> map = new HashMap<>();
            
            for (Text line : values) {
                String[] tokens = Recommend.DELIMITER.split(line.toString());
                String itemID = tokens[0];
                Double score = Double.parseDouble(tokens[1]);
                
                 if (map.containsKey(itemID)) {
                     map.put(itemID, map.get(itemID) + score);
                 } else {
                     map.put(itemID, score);
                 }
            }
            
            Iterator<String> iter = map.keySet().iterator();
            while (iter.hasNext()) {
                String itemID = iter.next();
                double score = map.get(itemID);
                Text v = new Text(itemID + "," + score);
                context.write(key, v);
            }
        }
    }

    public static void run(Map<String, String> path) throws IOException, InterruptedException, ClassNotFoundException {
        JobConf conf = Recommend.config();

        String input = path.get("Step6Input");
        String output = path.get("Step6Output");

        HdfsDAO hdfs = new HdfsDAO(Recommend.HDFS, conf);
        hdfs.rmr(output);

        Job job = new Job(conf);
        job.setJarByClass(Step4_Update2.class);

        job.setOutputKeyClass(Text.class);
        job.setOutputValueClass(Text.class);

        job.setMapperClass(Step4_Update2.Step4_RecommendMapper.class);
        job.setReducerClass(Step4_Update2.Step4_RecommendReducer.class);

        job.setInputFormatClass(TextInputFormat.class);
        job.setOutputFormatClass(TextOutputFormat.class);

        FileInputFormat.setInputPaths(job, new Path(input));
        FileOutputFormat.setOutputPath(job, new Path(output));

        job.waitForCompletion(true);
    }
}

辅助工具类：HDFS 操作封装

为了方便管理中间文件，我们封装了一个简单的 HDFS 工具类，支持创建目录、删除、重命名等操作。

package recommand;

import java.io.IOException;
import java.net.URI;

import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.fs.FSDataInputStream;
import org.apache.hadoop.fs.FSDataOutputStream;
import org.apache.hadoop.fs.FileStatus;
import org.apache.hadoop.fs.FileSystem;
import org.apache.hadoop.fs.Path;
import org.apache.hadoop.io.IOUtils;
import org.apache.hadoop.mapred.JobConf;

public class HdfsDAO {

    private static final String HDFS = "hdfs://115.28.167.22:9000/";

    public HdfsDAO(Configuration conf) {
        this(HDFS, conf);
    }

    public HdfsDAO(String hdfs, Configuration conf) {
        this.hdfsPath = hdfs;
        this.conf = conf;
    }

    private String hdfsPath;
    private Configuration conf;

    public void mkdirs(String folder) throws IOException {
        Path path = new Path(folder);
        FileSystem fs = FileSystem.get(URI.create(hdfsPath), conf);
        if (!fs.exists(path)) {
            fs.mkdirs(path);
            System.out.println("Create: " + folder);
        }
        fs.close();
    }

    public void rmr(String folder) throws IOException {
        Path path = new Path(folder);
        FileSystem fs = FileSystem.get(URI.create(hdfsPath), conf);
        fs.deleteOnExit(path);
        System.out.println("Delete: " + folder);
        fs.close();
    }

    public void rename(String src, String dst) throws IOException {
        Path name1 = new Path(src);
        Path name2 = new Path(dst);
        FileSystem fs = FileSystem.get(URI.create(hdfsPath), conf);
        fs.rename(name1, name2);
        System.out.println("Rename: from " + src + " to " + dst);
        fs.close();
    }

    public void ls(String folder) throws IOException {
        Path path = new Path(folder);
        FileSystem fs = FileSystem.get(URI.create(hdfsPath), conf);
        FileStatus[] list = fs.listStatus(path);
        System.out.println("ls: " + folder);
        for (FileStatus f : list) {
            System.out.printf("name: %s, folder: %s, size: %d\n", f.getPath(), f.isDir(), f.getLen());
        }
        fs.close();
    }

    public void createFile(String file, String content) throws IOException {
        FileSystem fs = FileSystem.get(URI.create(hdfsPath), conf);
        byte[] buff = content.getBytes();
        FSDataOutputStream os = null;
        try {
            os = fs.create(new Path(file));
            os.write(buff, 0, buff.length);
            System.out.println("Create: " + file);
        } finally {
            if (os != null) os.close();
        }
        fs.close();
    }

    public void copyFile(String local, String remote) throws IOException {
        FileSystem fs = FileSystem.get(URI.create(hdfsPath), conf);
        fs.copyFromLocalFile(new Path(local), new Path(remote));
        System.out.println("copy from: " + local + " to " + remote);
        fs.close();
    }

    public void download(String remote, String local) throws IOException {
        Path path = new Path(remote);
        FileSystem fs = FileSystem.get(URI.create(hdfsPath), conf);
        fs.copyToLocalFile(path, new Path(local));
        System.out.println("download: from" + remote + " to " + local);
        fs.close();
    }

    public void cat(String remoteFile) throws IOException {
        Path path = new Path(remoteFile);
        FileSystem fs = FileSystem.get(URI.create(hdfsPath), conf);
        FSDataInputStream fsdis = null;
        System.out.println("cat: " + remoteFile);
        try {
            fsdis = fs.open(path);
            IOUtils.copyBytes(fsdis, System.out, 4096, false);
        } finally {
            IOUtils.closeStream(fsdis);
            fs.close();
        }
    }
}

总结

这套代码实现了一个标准的离线推荐链路。关键点在于多阶段的 Job 设计，避免了单步计算的数据倾斜问题。在实际生产中，建议引入 HBase 或 Redis 缓存热门物品，并对数据进行增量更新。注意调整 conf.setNumMapTasks 和 conf.setNumReduceTasks 以优化集群资源利用率。