How to find common elements among two array column

问题:

I have two comma-separated string columns (sourceAuthors and targetAuthors).

val df = Seq(
  ("Author1,Author2,Author3","Author2,Author3,Author1")
).toDF("source","target")

I'd like to add another column nCommonAuthors with the number of common Authors.

I've tried doing this by this way:

def myUDF = udf { (s1: String, s2: String) =>
  s1.split(",")
  s2.split(",")
  s1.intersect(s2).length
}
val newDF = myDF.withColumn("nCommonAuthors", myUDF($"source", $"target"))

I get the following error :

Exception in thread "main" java.lang.UnsupportedOperationException: Schema for type Unit is not supported

Any idea why I get this error? How to find the common elements among two columns?

回答1:

Based on SCouto answer, I give you the complete solution that worked for me:

  def myUDF: UserDefinedFunction = udf(
(s1: String, s2: String) => {
  val splitted1 = s1.split(",")
  val splitted2 = s2.split(",")
  splitted1.intersect(splitted2).length
})

  val spark = SparkSession.builder().master("local").getOrCreate()

  import spark.implicits._

  val df = Seq(("Author1,Author2,Author3","Author2,Author3,Author1")).toDF("source","target")

  df.show(false)

+-----------------------+-----------------------+
|source                 |target                 |
+-----------------------+-----------------------+
|Author1,Author2,Author3|Author2,Author3,Author1|
+-----------------------+-----------------------+

  val newDF: DataFrame = df.withColumn("nCommonAuthors", myUDF('source,'target))

  newDF.show(false)

+-----------------------+-----------------------+--------------+
|source                 |target                 |nCommonAuthors|
+-----------------------+-----------------------+--------------+
|Author1,Author2,Author3|Author2,Author3,Author1|3             |
+-----------------------+-----------------------+--------------+

回答2:

Unless I misunderstood your problem, there are standard functions that can help you (so you don't have to write a UDF), i.e. split and array_intersect.

Given the following dataset:

val df = Seq(("Author1,Author2,Author3","Author2,Author3"))
  .toDF("source","target")
scala> df.show(false)
+-----------------------+---------------+
|source                 |target         |
+-----------------------+---------------+
|Author1,Author2,Author3|Author2,Author3|
+-----------------------+---------------+

You could write the following structured query:

val intersect = array_intersect(split('source, ","), split('target, ","))
val solution = df.select(intersect as "common_elements")
scala> solution.show(false)
+------------------+
|common_elements   |
+------------------+
|[Author2, Author3]|
+------------------+

回答3:

That error means that your udf is returning unit ( no return at all, as void un Java )

Try this. You are applying the intersect over the original s1 and S2 rather than over the splitted ones.

def myUDF = udf((s1: String, s2: String) =>{

  val splitted1 = s1.split(",")


  val splitted2= s2.split(",")


splitted1.intersect(splitted2).length

} )