Evaluates a client-side model built on top of the Universal Sentence Encoder to detect hateful content selected for the analysis