Research On Intelligent Question Answering Technology Of Ship Navigation Based On External Knowledge

The intelligent era is coming,and convenience,speed and intelligence have become the objects sought after by task people.In this era,our pursuit of efficient intelligence has never stopped.Intelligent question answering system is an important exploration and development direction of intelligent era.The ultimate goal of intelligent question answering system is seamless human-machine interaction.We expect that intelligent question-and-answer and intelligent conversation tasks will eventually allow computers to converse and answer questions as freely as humans.But all aspects of intelligent question answering difficult to achieve under the existing technology,after all,the man himself is to rely on for a long time continuous learning progress and save a lot of learning and memory to achieve interpersonal question and answer,and in the face of such a large number of computer knowledge,although can realize the storage of knowledge,but the right call such a large amount of information in the right one or a few really difficult to achieve.If we change our thinking and transform such a huge amount of memory into a certain field,it is just like the division of labor of human beings.People who are good at different fields are assigned to become experts in a certain field.Such an expert doesn’t have to memorize a lot of tedious and useless knowledge,but only studies his or her own field,which can be very helpful in answering questions about that particular field.The same is true for computers,so the research of intelligent question answering of ship navigation in this paper can realize efficient intelligent question answering in the field of ship navigation.In this paper we propose data sets and question answering models for intelligent question answering task of ship navigation:First,the question answering datasets is divided into question answering datasets and visual question answering datasets.The plain question answering datasets includes questions,answers,and facts.The visual question answering datasets includes questions,answers,images,and facts.The facts are in the form of knowledge graphs.the follow-up question-answering method research is based on these two data sets.Secondly,this paper proposes a bi-attention network(BANet)based on Tansformer for the pure question answering datasets.With attention as the main method,the model can deeply understand the problem and factual information,and complete efficient pure text q&A.Thirdly,in order to study visual question answering,this paper proposes a three-layer attention network applicable to the public data set FVQA.The network takes the input information of question,image and fact as three different modes,completes the interaction of the three modes of information with the multi-layer attention interaction network,and collects the useful information centrally to predict the answer.Fourth,based on the research of visual question answering,We improve the visual question answering algorithm based on knowledge to make it more suitable for the data set of ship navigation visual question answering proposed in this paper.