Utilizing Large Language Models to Generate Synthetic Data to Increase the Performance of BERT-Based Neural Networks