Question

我正试图从多个文档中生成一个术语 - 文档矩阵。我可以从已经创建的矩阵运行LDA模型，现在我需要退一步。我试图实现一个简单的term-doc矩阵，但现在我被困了。我做的是：

//GETS ALL FILES FROM INPUT PATH
JavaPairRDD<String, String> doc_words = context.wholeTextFiles(input_path);

//SPLIT BY " "
JavaPairRDD<String, String> tokenized = doc_words.flatMapValues(Preprocessing_DocumentTermMatrix.WORDS_EXTRACTOR);

//SEE METHOD WORDS_MAPPER.
JavaRDD<Tuple2<Tuple2<String, String>, Integer>> rdd = tokenized.flatMap(WORDS_MAPPER);


//METHOD WORDS_MAPPER
public static final FlatMapFunction<Tuple2<String, String>, Tuple2<Tuple2<String, String>, Integer>> WORDS_MAPPER = new FlatMapFunction<Tuple2<String, String>, Tuple2<Tuple2<String, String>, Integer>>() {

    public Iterable<Tuple2<Tuple2<String, String>, Integer>> call(Tuple2<String, String> stringIntegerTuple2) throws Exception {
        return Arrays.asList(new Tuple2<Tuple2<String, String>, Integer>(new Tuple2<String,String>(stringIntegerTuple2._1(), stringIntegerTuple2._2()), 1)); 
    } 
};

所以，这个函数给我一个这样的结果：

((DOC_0, TERM0), 1)
((DOC_0, TERM0), 1)
((DOC_0, TERM1), 1)
((DOC_1, TERM0), 1)
((DOC_1, TERM2), 1)

我想这很好，但现在我需要减少它并提取这样的输出：

(DOC_0, (TERM0, 2), (TERM1, 1))
(DOC_1, (TERM0, 1), (TERM2, 1))

我已经尝试了很多东西并且无法得到它......有人可以帮助我吗？

Answer 1

这是解决方案：

JavaPairRDD<String, Iterable<Tuple2<String, Integer>>> newrdd = JavaPairRDD.fromJavaRDD(rdd).reduceByKey((a, b) -> a + b)
                .mapToPair(t -> new Tuple2<>(t._1._1, new Tuple2<>(t._1._2, t._2))).groupByKey();

如何使用Spark从多个文档中获取term-document矩阵？

1 个答案: