In some consumer markets, prices are determined by the limited availability of goods and customers choose from a small set of options. The outcome is often determined by simple tradeoffs between the most critical attributes.

Here, the authors discuss a model for customer preferences in settings with many users and observable choices. The Gaussian process scalable preference model via Kronecker factorization (GaSPK) balances accurate choice predictions and scalability, improving upon previous approaches with good performance and limited scalability, or good scalability but mediocre performance. Utilizing the reliance of users on a small number of attributes, GaSPK employs dimensionality reduction techniques with an iterative scheme that identifies specific tradeoffs for different user populations.

To validate their approach, the authors use datasets on tariffs in smart electric grids, car purchases, and voter preferences among political parties in the United Kingdom (UK). They compare GaSPK against prior work, yielding a good balance between prediction performance and scalability as long as user decisions are based on a relatively small number of attributes with limited values.

Companies offering goods and services in such environments may be able to optimize their pricing schemes via GaSPK by creating autonomous decision-making models for consumer choice. Consumers, however, typically do not have access to the training data required and are likely to face situations where prices change continually, offerings are difficult to compare, and crucial information is hard to obtain. While the authors suggest additional domains to be explored, it would be interesting to see if their approach can be used for the benefit of consumers as well.