추천 시스템 R패키지 비교 연구

2018/06/14

추천 R패키지 속도 및 성능 비교 연구 결과입니다.

  • 데이터셋 : MovieLense (100만행)
  • 컴퓨터 사양 : 7i, 32Gb RAM


패키지 목록

package name package description
Myrrix Real-Time, Scalable Clustering and Recommender System, Evolved from Apache Mahout
recommenderlab Lab for Developing and Testing Recommender Algorithms
recosystem Recommender System using Matrix Factorization
rrecsys Environment for Evaluating Recommender Systems
slimrec Sparse Linear Method to Predict Ratings and Top-N Recommendations


패키지 성능 비교

package name algorithm time(min) RMSE
recommenderlab Most Popular 4.27 0.9725
  User-Based CF 5.03 1.0464
  Item-based CF 7.11 1.5074
  SVD 5.52 1.0204
  Funk SVD 13.91 0.9106
  Random 3.49 1.3832
  ALS 13.14 0.9032
rrecsys itemAverage 7.37 0.9614
  userAverage 6.95 1.0140
  globalAverage 6.22 1.0913
  IBKNN 7.53 1.0853
  UBKNN 37.49 1.0196
  FunkSVD 31.36 1.0811
  SlopeOne 15.48 0.9028
recosystem Matrix Factorization 0.68 0.8529
slimrec Sparse Linear Method 25.52 2.2196
SVDApproximation SVDApproximation 4.92 0.9313
SmartCat-labs’s Git R code ibcf 1.76 0.8859
  ubcf 1.74 0.8564


전체 분석 코드


-->