forked from D-Net/dnet-hadoop
176 lines
8.4 KiB
Java
176 lines
8.4 KiB
Java
package eu.dnetlib.dhp.graph;
|
|
|
|
import com.fasterxml.jackson.annotation.JsonInclude;
|
|
import com.fasterxml.jackson.core.JsonProcessingException;
|
|
import com.fasterxml.jackson.databind.ObjectMapper;
|
|
import com.google.common.collect.Iterables;
|
|
import com.google.common.collect.Lists;
|
|
import com.jayway.jsonpath.DocumentContext;
|
|
import com.jayway.jsonpath.JsonPath;
|
|
import eu.dnetlib.dhp.schema.oaf.*;
|
|
import net.minidev.json.JSONArray;
|
|
import org.apache.hadoop.io.Text;
|
|
import org.apache.hadoop.io.compress.GzipCodec;
|
|
import org.apache.spark.api.java.JavaPairRDD;
|
|
import org.apache.spark.api.java.JavaRDD;
|
|
import org.apache.spark.api.java.JavaSparkContext;
|
|
import org.apache.spark.api.java.function.PairFunction;
|
|
import org.apache.spark.sql.SparkSession;
|
|
import scala.Tuple2;
|
|
|
|
import java.io.IOException;
|
|
import java.io.Serializable;
|
|
import java.util.List;
|
|
import java.util.stream.Collectors;
|
|
|
|
/**
|
|
* Joins the graph nodes by resolving the links of distance = 1 to create an adjacency list of linked objects.
|
|
* The operation considers all the entity types (publication, dataset, software, ORP, project, datasource, organization,
|
|
* and all the possible relationships (similarity links produced by the Dedup process are excluded).
|
|
*
|
|
* The operation is implemented creating the union between the entity types (E), joined by the relationships (R), and again
|
|
* by E, finally grouped by E.id;
|
|
*
|
|
* Different manipulations of the E and R sets are introduced to reduce the complexity of the operation
|
|
* 1) treat the object payload as string, extracting only the necessary information beforehand using json path,
|
|
* it seems that deserializing it with jackson's object mapper has higher memory footprint.
|
|
*
|
|
* 2) only consider rels that are not virtually deleted ($.dataInfo.deletedbyinference == false)
|
|
* 3) we only need a subset of fields from the related entities, so we introduce a distinction between E_source = S
|
|
* and E_target = T. Objects in T are heavily pruned by all the unnecessary information
|
|
*
|
|
* 4) perform the join as (((T join R) union S) groupby S.id) yield S -> [ <T, R> ]
|
|
*/
|
|
public class GraphJoiner implements Serializable {
|
|
|
|
public static final int MAX_RELS = 100;
|
|
|
|
public void join(final SparkSession spark, final String inputPath, final String hiveDbName, final String outPath) {
|
|
|
|
final JavaSparkContext sc = new JavaSparkContext(spark.sparkContext());
|
|
|
|
// read each entity
|
|
JavaPairRDD<String, TypedRow> datasource = readPathEntity(sc, inputPath, "datasource");
|
|
JavaPairRDD<String, TypedRow> organization = readPathEntity(sc, inputPath, "organization");
|
|
JavaPairRDD<String, TypedRow> project = readPathEntity(sc, inputPath, "project");
|
|
JavaPairRDD<String, TypedRow> dataset = readPathEntity(sc, inputPath, "dataset");
|
|
JavaPairRDD<String, TypedRow> otherresearchproduct = readPathEntity(sc, inputPath, "otherresearchproduct");
|
|
JavaPairRDD<String, TypedRow> software = readPathEntity(sc, inputPath, "software");
|
|
JavaPairRDD<String, TypedRow> publication = readPathEntity(sc, inputPath, "publication");
|
|
|
|
// create the union between all the entities
|
|
final String entitiesPath = outPath + "/entities";
|
|
datasource
|
|
.union(organization)
|
|
.union(project)
|
|
.union(dataset)
|
|
.union(otherresearchproduct)
|
|
.union(software)
|
|
.union(publication)
|
|
.map(e -> new EntityRelEntity().setSource(e._2()))
|
|
.map(MappingUtils::serialize)
|
|
.saveAsTextFile(entitiesPath, GzipCodec.class);
|
|
|
|
JavaPairRDD<String, EntityRelEntity> entities = sc.textFile(entitiesPath)
|
|
.map(t -> new ObjectMapper().readValue(t, EntityRelEntity.class))
|
|
.mapToPair(t -> new Tuple2<>(t.getSource().getSourceId(), t));
|
|
|
|
// reads the relationships
|
|
final JavaPairRDD<String, EntityRelEntity> relation = readPathRelation(sc, inputPath)
|
|
.filter(r -> !r.getDeleted()) //only consider those that are not virtually deleted
|
|
.map(p -> new EntityRelEntity().setRelation(p))
|
|
.mapToPair(p -> new Tuple2<>(p.getRelation().getSourceId(), p))
|
|
.groupByKey()
|
|
.map(p -> Iterables.limit(p._2(), MAX_RELS))
|
|
.flatMap(p -> p.iterator())
|
|
.mapToPair(p -> new Tuple2<>(p.getRelation().getTargetId(), p));
|
|
|
|
final String joinByTargetPath = outPath + "/join_by_target";
|
|
relation
|
|
.join(entities
|
|
.filter(e -> !e._2().getSource().getDeleted())
|
|
.mapToPair(e -> new Tuple2<>(e._1(), MappingUtils.pruneModel(e._2()))))
|
|
.map(s -> new EntityRelEntity()
|
|
.setRelation(s._2()._1().getRelation())
|
|
.setTarget(s._2()._2().getSource()))
|
|
.map(MappingUtils::serialize)
|
|
.saveAsTextFile(joinByTargetPath, GzipCodec.class);
|
|
|
|
JavaPairRDD<String, EntityRelEntity> bySource = sc.textFile(joinByTargetPath)
|
|
.map(t -> new ObjectMapper().readValue(t, EntityRelEntity.class))
|
|
.mapToPair(t -> new Tuple2<>(t.getRelation().getSourceId(), t));
|
|
|
|
final String linkedEntitiesPath = outPath + "/linked_entities";
|
|
entities
|
|
.union(bySource)
|
|
.groupByKey() // by source id
|
|
.map(GraphJoiner::asLinkedEntityWrapper)
|
|
.map(MappingUtils::serialize)
|
|
.saveAsTextFile(linkedEntitiesPath, GzipCodec.class);
|
|
}
|
|
|
|
/**
|
|
* Reads a set of eu.dnetlib.dhp.schema.oaf.OafEntity objects from a sequence file <className, entity json serialization>,
|
|
* extracts necessary information using json path, wraps the oaf object in a eu.dnetlib.dhp.graph.TypedRow
|
|
* @param sc
|
|
* @param inputPath
|
|
* @param type
|
|
* @return the JavaPairRDD<String, TypedRow> indexed by entity identifier
|
|
*/
|
|
private JavaPairRDD<String, TypedRow> readPathEntity(final JavaSparkContext sc, final String inputPath, final String type) {
|
|
return sc.sequenceFile(inputPath + "/" + type, Text.class, Text.class)
|
|
.mapToPair((PairFunction<Tuple2<Text, Text>, String, TypedRow>) item -> {
|
|
|
|
final String json = item._2().toString();
|
|
final String id = JsonPath.read(json, "$.id");
|
|
return new Tuple2<>(id, new TypedRow()
|
|
.setSourceId(id)
|
|
.setDeleted(JsonPath.read(json, "$.dataInfo.deletedbyinference"))
|
|
.setType(type)
|
|
.setOaf(json));
|
|
});
|
|
}
|
|
|
|
/**
|
|
* Reads a set of eu.dnetlib.dhp.schema.oaf.Relation objects from a sequence file <className, relation json serialization>,
|
|
* extracts necessary information using json path, wraps the oaf object in a eu.dnetlib.dhp.graph.TypedRow
|
|
* @param sc
|
|
* @param inputPath
|
|
* @return the JavaRDD<TypedRow> containing all the relationships
|
|
*/
|
|
private JavaRDD<TypedRow> readPathRelation(final JavaSparkContext sc, final String inputPath) {
|
|
return sc.sequenceFile(inputPath + "/relation", Text.class, Text.class)
|
|
.map(item -> {
|
|
final String s = item._2().toString();
|
|
final DocumentContext json = JsonPath.parse(s);
|
|
return new TypedRow()
|
|
.setSourceId(json.read("$.source"))
|
|
.setTargetId(json.read("$.target"))
|
|
.setDeleted(json.read("$.dataInfo.deletedbyinference"))
|
|
.setType("relation")
|
|
.setOaf(s);
|
|
});
|
|
}
|
|
|
|
private static LinkedEntityWrapper asLinkedEntityWrapper(Tuple2<String, Iterable<EntityRelEntity>> p) {
|
|
final LinkedEntityWrapper e = new LinkedEntityWrapper();
|
|
final List<TupleWrapper> links = Lists.newArrayList();
|
|
for (EntityRelEntity rel : p._2()) {
|
|
if (rel.hasMainEntity() & e.getEntity() == null) {
|
|
e.setEntity(rel.getSource());
|
|
}
|
|
if (rel.hasRelatedEntity()) {
|
|
links.add(new TupleWrapper()
|
|
.setRelation(rel.getRelation())
|
|
.setTarget(rel.getTarget()));
|
|
}
|
|
}
|
|
e.setLinks(links);
|
|
if (e.getEntity() == null) {
|
|
throw new IllegalStateException("missing main entity on '" + p._1() + "'");
|
|
}
|
|
return e;
|
|
}
|
|
|
|
}
|