We propose a latent variable model for paired comparisons in the context of Basketball scores. Based on the model proposed by [1], our model assumes multi-dimensional latent skills decomposed into offensive and defensive components as well as team- and conference-specific skills interacting with each other through an inner product model. We propose two inference approaches: maximum likelihood estimation (MLE) and mean-field variational inference (VI). Applied to the 2004- 2017 NCAA Men’s Basketball season’s, the MLE approach yields adequate and interpretable results, but the VI approach was not as successful. Based on the MLE inference, we study the relationship between teams and conferences, we investigate league-wide trends over time and we produce rankings of teams and conferences.