Generalized metrics for evaluating recommendation algorithms as a criterion for optimizing hyperparameters

This paper considers an approach to optimizing hyperparameters of recommendation algorithms using an integral assessment that combines several metrics into four key subindexes: accuracy, ranking, diversity, and resource intensity. This method allows for a more balanced tuning of models, ensuring improved quality of recommendations without loss in individual characteristics. Unlike adjusting for one metric, which can worsen the rest of the parameters, the proposed approach takes into account the mutual dependencies between the metrics. It is shown that different algorithms react differently to the chosen optimization strategy. The developed methodology can be applied not only in recommendation systems, but also in other multi—criteria optimization tasks, such as machine learning, financial modeling, and engineering applications.