In BPR paper BPR: Bayesian personalized ranking from implicit feedback (Steffen Rendle 2009)
$$ \prod_{u \in U} p\left(>_{u} | \Theta\right)=\prod_{(u, i, j) \in U \times I \times I} p\left(i>_{u} j | \Theta\right)^{\delta\left((u, i, j) \in D_{S}\right) } \cdot\left(1-p\left(i>_{u} j | \Theta\right)\right)^{\delta\left((u, j, i) \notin D_{S}\right)} $$
Why the power of second term should be $\delta\left((u, j, i) \notin D_{S}\right)$ ?Shouldn't it be $\delta\left((u, i, j) \notin D_{S}\right)$? It is obvious that $(u, i, j) \in D_{S}$ implies $(u, j, i) \notin D_{S}$.
And how can it be reduced to be
$$ \prod_{u \in U} p\left(>_{u} | \Theta\right)=\prod_{(u, i, j) \in D_{S}} p\left(i>_{u} j | \Theta\right) $$
And $$ \begin{aligned} \mathrm{BPR}-\mathrm{OPT} & :=\ln p\left(\Theta |>_{u}\right) \\ &=\ln p\left(>_{u} | \Theta\right) p(\Theta) \\ &=\ln \prod_{(u, i, j) \in D_{S}} \sigma\left(\hat{x}_{u i j}\right) p(\Theta) \\ &=\sum_{(u, i, j) \in D_{S}} \ln \sigma\left(\hat{x}_{u i j}\right)+\ln p(\Theta) \\ &=\sum_{(u, i, j) \in D_{S}} \ln \sigma\left(\hat{x}_{u i j}\right)-\lambda_{\Theta}\|\Theta\|^{2} \end{aligned} $$
Does this formula in the last equation miss $\ln p(\Theta)$
Any help will be appreciated!!!


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.