Enhancing Cluster-Based Topic Models Through Parametric Dimensionality Reduction with Transformer-Encoders