Apply deep learning to improve the question analysis model in the Vietnamese question answering system

Question answering (QA) system nowadays is quite popular for automated purposes, the meaning analysis of question plays an important role, directly affecting accuracy system. In this article, we propose improvement question-answering models by adding more specific steps, including contextual characteristic analysis, pos-tag and question-type built on deep learning network architecture. Weights extracted words through steps are combined with best matching 25 (BM25) algorithm to find relevant paragraph text incorporated into QA model least noisy answer. The dataset step consists 19,339 labeled questions covering a variety topics. Results train data set related regulations Industrial University Ho Chi Minh City. It includes 17,405 pairs answers training 1,600 test set, where robustly optimized BERT pre-training approach (RoBERTa) has F1-score 74%. improved significantly. For long complex questions, mode weights correctly provided based question’s contents.