This paper proposes a new Deep Learning-based Query Expansion (DL-QE) framework dedicated to improving the effectiveness of next generation information retrieval systems. The DL-QE framework utilizes BERT, Word2Vec, and reinforcement learning to semantically enhance user queries to improve retrieval accuracy. We demonstrate using the TREC-3, CRAN, and MED benchmark datasets that DL-QE is a superior method than traditional approaches to query expansion. For example, DL-QE achieves a Precision@10 of 0.80, Recall of 0.85, F1-Score of 0.83, and MAP of 0.80 on the CRAN dataset. On the MED dataset, we obtain a Precision@10 score of 0.83, Recall of 0.87, and an F1-Score of 0.85. Compared to baseline BM25 and VSM models, where all metrics were below 0.66, we made significant improvement across all metrics, and in our understanding of the semantics of words. In addition, transformer-based neural ranking models improve relevance to users by determining which documents are similar to the user's query. The DL-QE framework demonstrates compatibility and scalability for modelling queries in various information environments. As a result, the DL-QE framework can facilitate dynamic, real-time reformulating of queries.