The Research on Re-ranking Algorithm for FAQ-based Systems in the Petroleum Domain

In an FAQ-based system, precise information retrieval and recall are key to enhancing user experience. However, traditional retrieval methods struggle with professional terminology and complex semantic matching, making it difficult to achieve high-precision QA. This study explores the application of reranking algorithms in an oil industry FAQ system, aiming to optimize the ranking of recalled candidate answers and improve QA matching accuracy. We employ a deep learning-based dense retrieval (DPR) model for initial recall and integrate a pretrained reranking model (BAAI/bge-reranker) to refine the retrieved results. Experimental results demonstrate that incorporating a reranking algorithm significantly improves recall precision. Additionally, by fine-tuning the model with domain-specific petroleum data, the reranking task achieves better performance within the field, further validating the value of reranking algorithms in enhancing professional knowledge QA matching.