Starting from:

$30

Assignment 6 Task: PageRank

Web Search and Sense-Making
Assignment 6

Task: PageRank
Introduction:
In this assignment, we will implement the PageRank algorithm on the Wikipedia data.
Requirements:
100GB free disk space in your machine.
Instructions:
Write a PageRank.scala file to calculate and sort the PageRank scores for the English
Wikipedia pages, by taking the following steps:
1. Read in the output files of your last assignment, which contains the link graph in Wikipedia.
The format should be:
• One page per line
• In each line, you have the title of a page and a list of the titles of the outlinks in the page
• Each outline title is inside [[]], and separated by a tab “\t”.
• The title and the list of links is separated by a delimiter. We recommended “\t”.
2. Create two pair RDDs from the input file. One pair RDD holds the page and its outlinks and
another pair RDD holds the page and its PageRank score.
3. Implement the PageRank algorithm. Calculate PageRank scores for all the pages in the
input file. Use ITERATION = 20.
4. Sort the PageRank scores for the pages in the descending order.
5. Save the title of a page and its PageRank scores into file. The format is like:
[[Serie A]] 4.798429823418915
[[United States dollar]] 4.796979625049858
[[Iraq]] 4.764439498823581
[[Columbia Pictures]] 4.762980165170773
[[mixed martial arts]] 4.756275511978413
[[Rio de Janeiro]] 4.7528282896086465
[[People (magazine)]] 4.7463648409870185
6. You are welcome to use the following code template:
import scala.util.matching.Regex
import org.apache.spark.SparkConf
import org.apache.spark.SparkContext
import org.apache.spark.SparkContext._
import org.apache.hadoop.io.compress.GzipCodec
COSC 589 - Web Search and Sense-Making
object PageRank {
def main(args: Array[String]) {
val sparkConf = new SparkConf().setAppName(“PageRank”)
val sc = new SparkContext(sparkConf)
val input = sc.textFile(“./linkgraph/*.gz”) // your output directory from the last
assignment
val links = // Load RDD of (page title, links) pairs
val ranks = // Load RDD of (page title, rank) pairs

val ITERATION = 20
// Implement your PageRank algorithm according to the notes


// Sort pages by their PageRank scores
ranks.sortBy …
// save the page title and pagerank scores in compressed format (save your disk
space). Using “\t” as the delimiter.
ranks.map(r = r._1 + "\t" + r._2).saveAsTextFile("./pageranks", classOf[GzipCodec])
}
}
What to Submit:
- Your code
- Screen captures of the beginning of your pageranked pages, in descending order (e.g. the
first 20 lines on the screen. Hint: Use ‘gunzip part-00000.gz” to unzip, then view the
documents and screen capture)
Bonus: (20% of the entire grade of this homework)
Output pages and their PageRank scores for only Persons.
What to Submit:
- Your code
- Screen captures of the beginning of your pageranked pages, in descending order (e.g. the
first 20 lines on the screen. Hint: Use ‘gunzip part-00000.gz” to unzip, then view the
documents and screen capture)
What NOT to Submit:
- Your input or output files
Where to submit:
COSC 589 - Web Search and Sense-Making
- Canvas

More products