Complex Question Decomposition Method: Based on Fine-grained Named Entity Recognition and Domain Knowledge

With the development of the intelligent question answering technology, and the decomposition of complex problems has gradually become a key technology for question answering systems. Aiming at the existing complex problem decomposition methods that cannot meet the design requirements of question answering systems in the field of weaponry. In this article, we proposed a complex question decomposition method based on fine-grained named entity recognition (FG-NER) and domain knowledge. By introducing the BERT-based Bi-LSTM- CRF model, it effectively solves the identification problem of weapon types, names and corresponding attributes in the complex question sentences with weaponry domain. Experimental results demonstrate that the model has an average Fi score of 91.54% on the test dataset. It can efficiently complete complex question decomposition tasks in the field of aircraft weaponry, and it can precisely to finish the questions-answers matching problem by combining the weaponry domain knowledge base.