According to eq (17) in the original WSDM paper, p(f|c) is based on |v_{c,f}| * std(f).
However, in librec AoBPR implementation here, p(f|c) is calculated with variance(f) directly. I'm wondering whether it is a bug in AoBPR implementation. Correct me please if I misunderstand something.
According to eq (17) in the original WSDM paper, p(f|c) is based on |v_{c,f}| * std(f).
However, in librec AoBPR implementation here, p(f|c) is calculated with variance(f) directly. I'm wondering whether it is a bug in AoBPR implementation. Correct me please if I misunderstand something.