The evolution of mathematical modeling and numerical simulation has provided increasingly accurate analyses of complex phenomena but increased computational cost. To address this challenge, reduction methods are being developed that maintain physical consistency while reducing complexity. In this work we study a class of structure-preserving collocation reduced order models, which integrate structural constraints to preserve fundamental properties. Particular attention is paid to conservation of entropy, which is essential to ensure physical consistency. It focuses on the Burgers equation, analyzing the construction of an entropy stable reduced model using proper orthogonal decomposition and validating its effectiveness with numerical tests. Finally, the hyper-reduced case is considered using a collocation strategy based on Nonnegative Least Squares, highlighting its limitations and possible future developments.