SimRank

1. Package name: package org.apache.spark.ml.recommendation
2. Class name: SimRank
3. Method name: computeSimilarity
4. Input: Dataset[_] that stores user search and product usage statistics. At least two columns are included: users and products.

   Parameter	Value Type	Description
   dataset	Dataset[_]	Dataset that stores user search and product usage statistics

5. Parameters optimized based on native algorithms

   Parameter	Value Type	Description
   damp	Double	Attenuation factor
   numIter	Int	Number of iterations
   userCol	String	User column
   itemCol	String	Product column

   It is recommended that damp be set to 0.6 and numIter to 5. Small attenuation factor may lead to low precision and large attenuation may lead to slow convergence. If the number of iterations is too small, the precision may decrease. If the number of iterations is too large, the computation may take a long time.

   

   Code API example:

   val simrank = new SimRank().setDamp(0.6).setNumIter(5).setUserCol("user").setItemCol("item")
   val simrankSimilarity = simrank.computeSimilarity(df)
   val userSim = simrankSimilarity.userSimilarity
   val itemSim = simrankSimilarity.itemSimilarity

6. Output: SimRankSimilarity, including the similarity between users and that between products. It is a case class that contains two variables.

   Parameter	Value Type	Description
   userSimilarity	DataFrame	Similarity between users
   itemSimilarity	DataFrame	Similarity between products

val simrank = new SimRank().setDamp(0.6).setNumIter(5).setUserCol("user").setItemCol("item")
val simrankSimilarity = simrank.computeSimilarity(df)
val userSim = simrankSimilarity.userSimilarity
val itemSim = simrankSimilarity.itemSimilarity