A Statistical Interpretation of Multi-Item Rating and Recommendation Problems

Ordinal user-provided ratings across multiple items are frequently encountered in both scientific and commercial applications. Whilst recommender systems are known to do well on these type of data from a predictive point of view, their typical reliance on large sample sizes and frequent lack of interpretability and uncertainty quantification limits their applicability in inferential problems. Taking a fully Bayesian approach, this article introduces a novel statistical method that is designed with interpretability and uncertainty quantification in mind. Whilst parametric assumptions ensure that the method is applicable to data with modest sample sizes, the model is simultaneously designed to remain flexible in order to handle a wide variety of situations. Model performance, i.e. parameter estimation and prediction, is shown by means of a simulation study, both on simulated data and against commonly used recommender systems on real data. These simulations indicate that the proposed method performs competitively. Finally, to illustrate the applicability of the proposed method on real life problems that are of interest to economists, the method is applied on speed dating data, where novel insights into the partner preference problem are obtained. An R package containing the proposed methodology can be found on https://CRAN.R-project.org/package=StatRec.