Building a Vietnamese question answering system based on knowledge graph and distributed CNN

Question answering system (QAS) can be applied everywhere such as in schools, hospitals, banks, e-commerce websites. A smart QAS that can replace people is what people expect. Therefore, there are a lot of studies to build, develop, and improve QAS. However, QAS used for low-resource languages like Vietnamese is still very limited. So, in this paper, we present a method for building Vietnamese QAS. Except for specific Vietnamese language processes, most of our solutions can also be applied to other languages. We build QAS based on knowledge graph (KG) and convolutional neural network (CNN). KG provides knowledge and deducing ability for QAS. CNN is used to classify questions in the natural language to identify the correct answer to a given question. Moreover, we also use distributed architecture to train the CNN model. On the other hands, we also propose a solution to speed up searching for answers in a large KG by partitioning and indexing KG by using the DM-Tree structure. Besides, we also present experimental results and evaluation results of our model using common metrics to prove the effectiveness of our solution.