# Prior and Likelihood Choices for Bayesian Matrix Factorisation on Small Datasets

arXiv: https://arxiv.org/abs/1712.00288 · HTML: https://mlsanity.com/paper/1712.00288/

## Key facts

- Cited by 4 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 1 Dec 2017.

## Built on by (most cited first, 4 of 4)

- [Flexible and Hierarchical Prior for Bayesian Nonnegative Matrix Factorization](https://mlsanity.com/paper/2205.11025/) (2022)
- [Bayesian Low-Rank Interpolative Decomposition for Complex Datasets](https://mlsanity.com/paper/2205.14825/) (2022)
- [Comparative Study of Inference Methods for Interpolative Decomposition](https://mlsanity.com/paper/2206.14542/) (2022)
- [Robust Bayesian Nonnegative Matrix Factorization with Implicit Regularizers](https://mlsanity.com/paper/2208.10053/) (2022)

## Abstract (arXiv)

In this paper, we study the effects of different prior and likelihood choices for Bayesian matrix factorisation, focusing on small datasets. These choices can greatly influence the predictive performance of the methods. We identify four groups of approaches: Gaussian-likelihood with real-valued priors, nonnegative priors, semi-nonnegative models, and finally Poisson-likelihood approaches. For each group we review several models from the literature, considering sixteen in total, and discuss the relations between different priors and matrix norms. We extensively compare these methods on eight real-world datasets across three application areas, giving both inter- and intra-group comparisons. We measure convergence runtime speed, cross-validation performance, sparse and noisy prediction performance, and model selection robustness. We offer several insights into the trade-offs between prior and likelihood choices for Bayesian matrix factorisation on small datasets - such as that Poisson models give poor predictions, and that nonnegative models are more constrained than real-valued ones.