dnet-hadoop/dhp-workflows/dhp-doiboost/src/main/java/eu/dnetlib/doiboost/mag/SparkPreProcessMAG.scala

package eu.dnetlib.doiboost.mag

import eu.dnetlib.dhp.application.ArgumentApplicationParser
import eu.dnetlib.dhp.schema.oaf.Publication
import eu.dnetlib.doiboost.DoiBoostMappingUtil.asField
import org.apache.commons.io.IOUtils
import org.apache.spark.SparkConf
import org.apache.spark.rdd.RDD
import org.apache.spark.sql.{Dataset, Encoder, Encoders, SaveMode, SparkSession}
import org.slf4j.{Logger, LoggerFactory}
import org.apache.spark.sql.functions._

import scala.collection.JavaConverters._

object SparkPreProcessMAG {


  def main(args: Array[String]): Unit = {

    val logger: Logger = LoggerFactory.getLogger(getClass)
    val conf: SparkConf = new SparkConf()
    val parser = new ArgumentApplicationParser(IOUtils.toString(getClass.getResourceAsStream("/eu/dnetlib/dhp/doiboost/mag/preprocess_mag_params.json")))
    parser.parseArgument(args)
    val spark: SparkSession =
      SparkSession
        .builder()
        .config(conf)
        .appName(getClass.getSimpleName)
        .master(parser.get("master")).getOrCreate()

    val sourcePath = parser.get("sourcePath")
    import spark.implicits._
    implicit val mapEncoderPubs: Encoder[Publication] = org.apache.spark.sql.Encoders.kryo[Publication]
    implicit val tupleForJoinEncoder = Encoders.tuple(Encoders.STRING, mapEncoderPubs)



    logger.info("Phase 1) make uninque DOI in Papers:")

    val d: Dataset[MagPapers] = spark.read.load(s"${parser.get("sourcePath")}/Papers").as[MagPapers]


    // Filtering Papers with DOI, and since for the same DOI we have multiple version of item with different PapersId we get the last one
    val result: RDD[MagPapers] = d.where(col("Doi").isNotNull).rdd.map { p: MagPapers => Tuple2(p.Doi, p) }.reduceByKey { case (p1: MagPapers, p2: MagPapers) =>
      var r = if (p1 == null) p2 else p1
      if (p1 != null && p2 != null) {
        if (p1.CreatedDate != null && p2.CreatedDate != null) {
          if (p1.CreatedDate.before(p2.CreatedDate))
            r = p1
          else
            r = p2
        } else {
          r = if (p1.CreatedDate == null) p2 else p1
        }
      }
      r
    }.map(_._2)

    val distinctPaper: Dataset[MagPapers] = spark.createDataset(result)
    distinctPaper.write.mode(SaveMode.Overwrite).save(s"${parser.get("targetPath")}/Papers_distinct")
    logger.info(s"Total number of element: ${result.count()}")

    logger.info("Phase 3) Group Author by PaperId")
    val authors = spark.read.load(s"$sourcePath/Authors").as[MagAuthor]

    val affiliation =spark.read.load(s"$sourcePath/Affiliations").as[MagAffiliation]

    val paperAuthorAffiliation =spark.read.load(s"$sourcePath/PaperAuthorAffiliations").as[MagPaperAuthorAffiliation]


    paperAuthorAffiliation.joinWith(authors, paperAuthorAffiliation("AuthorId").equalTo(authors("AuthorId")))
      .map{case (a:MagPaperAuthorAffiliation,b:MagAuthor )=>  (a.AffiliationId,MagPaperAuthorDenormalized(a.PaperId, b, null)) }
      .joinWith(affiliation, affiliation("AffiliationId").equalTo(col("_1")), "left")
      .map(s => {
          val mpa = s._1._2
          val af = s._2
          if (af!= null) {
            MagPaperAuthorDenormalized(mpa.PaperId, mpa.author, af.DisplayName)
          } else
            mpa
        }).groupBy("PaperId").agg(collect_list(struct($"author", $"affiliation")).as("authors"))
      .write.mode(SaveMode.Overwrite).save(s"${parser.get("targetPath")}/merge_step_1_paper_authors")



    logger.info("Phase 4) create First Version of publication Entity with Paper Journal and Authors")


    val journals = spark.read.load(s"$sourcePath/Journals").as[MagJournal]

    val papers =spark.read.load((s"${parser.get("targetPath")}/Papers_distinct")).as[MagPapers]

    val paperWithAuthors = spark.read.load(s"${parser.get("targetPath")}/merge_step_1_paper_authors").as[MagPaperWithAuthorList]



    val firstJoin =papers.joinWith(journals, papers("JournalId").equalTo(journals("JournalId")),"left")
    firstJoin.joinWith(paperWithAuthors, firstJoin("_1.PaperId").equalTo(paperWithAuthors("PaperId")), "left")
      .map { a: ((MagPapers, MagJournal), MagPaperWithAuthorList) => ConversionUtil.createOAFFromJournalAuthorPaper(a) }.write.mode(SaveMode.Overwrite).save(s"${parser.get("targetPath")}/merge_step_2")



    var magPubs:Dataset[(String,Publication)] = spark.read.load(s"${parser.get("targetPath")}/merge_step_2").as[Publication].map(p => (ConversionUtil.extractMagIdentifier(p.getOriginalId.asScala), p)).as[(String,Publication)]

    val paperUrlDataset = spark.read.load(s"$sourcePath/PaperUrls").as[MagPaperUrl].groupBy("PaperId").agg(collect_list(struct("sourceUrl")).as("instances")).as[MagUrl]


    logger.info("Phase 5) enrich publication with URL and Instances")

    magPubs.joinWith(paperUrlDataset, col("_1").equalTo(paperUrlDataset("PaperId")), "left")
      .map{a:((String,Publication), MagUrl) => ConversionUtil.addInstances((a._1._2, a._2))}
      .write.mode(SaveMode.Overwrite)
      .save(s"${parser.get("targetPath")}/merge_step_3")



    logger.info("Phase 6) Enrich Publication with description")
    val pa = spark.read.load(s"${parser.get("sourcePath")}/PaperAbstractsInvertedIndex").as[MagPaperAbstract]
    pa.map(ConversionUtil.transformPaperAbstract).write.mode(SaveMode.Overwrite).save(s"${parser.get("targetPath")}/PaperAbstract")

    val paperAbstract =spark.read.load((s"${parser.get("targetPath")}/PaperAbstract")).as[MagPaperAbstract]


    magPubs = spark.read.load(s"${parser.get("targetPath")}/merge_step_3").as[Publication].map(p => (ConversionUtil.extractMagIdentifier(p.getOriginalId.asScala), p)).as[(String,Publication)]

    magPubs.joinWith(paperAbstract,col("_1").equalTo(paperAbstract("PaperId")), "left").map(p=>
      {
        val pub = p._1._2
        val abst = p._2
        if (abst!= null) {
          pub.setDescription(List(asField(abst.IndexedAbstract)).asJava)
        }
        pub
        }
       ).write.mode(SaveMode.Overwrite).save(s"${parser.get("targetPath")}/merge_step_4")

  }


}