The rapid pace of digital transformation and the concurrent development of marine information technology have intensified the pressing need for seamless integration of ship data, information, and knowledge Q & A. Compared to traditional search engines, the knowledge graph-based Q & A system has attracted more and more attention from researchers due to its smarter, more efficient, and more accurate Q & A experience. This paper constructs a knowledge graph of ships and implements a ship knowledge question and answer system based on this graph. In order to better perform triple extraction from the knowledge text and recognize the intent of user questions, the paper proposes a novel BERT-CNN-Att named entity recognition model, which combines BERT, convolutional neural network, and Attention mechanism. Additionally, a BERT-BiLSTM relation extraction model is introduced, using BERT and Bidirectional Long Short-Term Memory (BiLSTM). Unlike traditional neural networks used for knowledge extraction, the named entity recognition model incorporate vocabulary feedback and enhancement mechanisms, enabling effective utilization of low-level representations for enriching semantic representations. Experimental results demonstrate the effectiveness of the proposed models in both named entity recognition and relation extraction tasks, showing significant improvements in processing speed. Furthermore, a detailed design for the Q & A system architecture is presented, leading to the successful development of an interactive ship knowledge question answering system based on the knowledge graph. Test results confirm that this system is capable of satisfying user requirements for ship-related knowledge queries.