Comparison results Approaches Amazon-200k Yelp-200k CIKM-Yelp CIKM-Douban 2.9656 2.5141 1.5323 0.7673 Regev (+60.0%) (+499%)(+277%) (+9.0%) 1.3462 1.7637 1.4342 0.7524 FMR (+119%)(+286%)(+2.8%) (+7.2%) 2.5368 2.3475 1.4891 0.7671 HeteRec (+532%)(+470%)(+256%) +9.0% 1.4603 1.1559 0.7216 Semrec HIN Based (+13.8%)‖(+4.2%) (+3.2%) Approaches MG11641.2581.1074 0.6985 HeteRec Yu et al. WSDM'14 SemRec shi et al., ClKM15 Factorize each meta-path Ensemble of original similarity Ensemble using the recovered matrices matrices based on different meta- Item-based CF User based ce Amazon- 200k Yelp-200k CIKM-Yelp CIKM-Douban Density 0.015% 0.024% 0.086% 0.630%Comparison Results • HeteRec [Yu et al., WSDM’14]: – Factorize each meta-path – Ensemble using the recovered matrices – Item-based CF 44 • SemRec [Shi et al., CIKM’15]: – Ensemble of original similarity matrices based on different meta￾paths – User based CF Traditional Approaches HIN Based Approaches