Generative Mongolian Question Answering System Based on Deep Learning

Question answering system is an important direction in artificial intelligence field. At present, the research on Mongolian Q&A system is still in its infancy, and its development is hindered by many factors, among which the biggest problem is that there is no publicly available Mongolian Q&A corpus. By collecting, screening, translating and proofreading existing Chinese Q&A corpus, this paper constructs Mongolian question answering corpus containing 80,000 question pairs. Based on the self-built Mongolian question answering corpus of 80,000 pairs, this paper uses Seq2seq +Attention and word2vec pretrained word vector for word embedding layer to study Mongolian question answering model. By comparison, it is found that Word2vec+ Bi-LSTM + LSTM-Attention model has the best performance in each evaluation index, and is more suitable for the realization of Mongolian question answering system. The study is a meaningful attempt.