A1917
Title: Deep ranking with heterogeneous effects
Authors: Shuxing Fang - The Hong Kong Polytechnic University (Hong Kong) [presenting]
Abstract: Classical latent-score ranking models often fail to distinguish objects' intrinsic scores from contextual effects, which are typically nonlinear and can dominate the observed outcomes. A semiparametric ranking framework is introduced in which the log-score of each object is modeled as the sum of a utility parameter and nonparametric covariate effects. Within this framework, model identifiability is established under mild regularity and connectivity conditions. For estimation, the covariate effects are approximated using a neural network and the parameters are estimated via maximum likelihood. Under random design assumptions, it is proven that the resulting estimator exists with high probability and non-asymptotic error bounds are derived that achieve minimax optimality for both the parametric and nonparametric components. Numerical experiments on both synthetic data and an ATP tennis dataset are conducted to support these findings.
