Commerce is increasingly moving toward an online space. With no way of inspecting goods directly users often rely on the experience of others when deciding on a parches. Reviews and seller reputation therefore play a key role in online sales. However, the sentiment expressed in reviews can often be unclear, and with the increasing number of user reviews manually addressing each review in order to provide feedback or improve service is often not a viable option. This work explores the potential of emerging natural language processing techniques in combination with the CatBoost classifier to interpret and correctly classify expressed sentiment in user review comments. As classifier performance depends on proper tuning, metaheuristics algorithms are employed to handle optimization and a modified version of efficient firefly algorithm optimizer is proposed. The introduced approach is tested on a real world data demonstrating promising outcomes with accuracy scores of 0.867944 for the best performing model optimized by the introduced optimizer.