def kcluster(rows, distance=:pearson, k=4)

    # それぞれのポイントの最小値と最大値を決める

    ranges = Array.new

    for i in 0...rows[0].length

      cols = Array.new

      for row in rows

        cols.push(row[i])

      end

      ranges.push([cols.min, cols.max])

    end

    # 重心をランダムにk個配置する

    clusters = Array.new

    k.times do

      tmp = Array.new

      for i in 0...rows[0].length

        tmp.push( rand()*(ranges[i].max-ranges[i].min)+ranges[i].min )

      end

      clusters.push(tmp)

    end

    # 重心を再計算しながら100回、もしくは変化が無くなるまで移動させる

    lastmatches = nil

    for t in 0...100

      puts 'Iteration ' + t.to_s

      bestmatches = Array.new # 空の配列作成

      k.times do bestmatches.push(Array.new) end

      # それぞれの行に対して、もっとも近い重心を探し出す

      for j in 0...rows.length

        row = rows[j]

        bestmatch = 0

        for i in 0...k

          d = self.method(distance).call(clusters[i],row) # 距離計算

          bestmatch = i if d < self.method(distance).call(clusters[bestmatch],row)

        end

        bestmatches[bestmatch].push(j)

      end

      # 結果が前回と同じであれば完了

      return bestmatches if bestmatches == lastmatches

      lastmatches = bestmatches

      # 重心をそのメンバーの平均に移動する

      for i in 0...k

        avgs = Array.new(rows[0].length, 0.0) # 初期値0.0の配列をいくつか作る

        if bestmatches[i].length > 0

          for rowid in bestmatches[i]

            for m in 0...rows[rowid].length

              avgs[m] += rows[rowid][m]

            end

          end

          for j in 0...avgs.length

            avgs[j] /= bestmatches[j].length if bestmatches[j] != nil

          end

          clusters[i] = avgs

        end

        #return bestmatches

      end

    end

  end

    v1 = [v1] if v1.class != Array

    v2 = [v2] if v2.class != Array

      f = node.feature.split(/,/)

    words = Array.new

    nouns.each{ |w|

      words.push(w) if w =~ /\w{2,}/ # 2文字以上

    }