Probabilistic Matrix Factoring (PMF) Model has been proven to be an effectiveand efficient method for recommendation. The most important parameter in PMFis the number of factors. Existing works usually choose this parameter heuristically or test several values by experiment and then choose a best one. Thesemethods are all very time expensive or can not archive the best performance. In this paper, we developed a novel method to choose this parameter proactively. Though there are many existing methods to choose the factors number in Principal Component Analysis, Factor Analysis, Singular Value Decomposition which inspire the PMF model,to our knowledge, there is no work studying how to determine the factor number in PMF proactively considering the difference between the traditional factor technique with PMF in recommendation.Experiments in two real world data sets show that our proposed method can choose the best parameter of the factor number for PMF model that archive the best performance.
展开▼
机译:An Exigency for Ice Core Studies to Determine Spatio-temporal Variability in Moisture Sources and Impact of Black Carbon - Mineral Aerosols on the Himalayan Glaciers