Research On Application Of High School Geography Automatic Answering System Based On Neural Network

"Question Answering based on Knowledge Retrieval and Inference Technology"is the key project of the National 863 program.The project needs to explore the level of human intelligence,aiming at the challenges brought about by semantic understanding in real scenes,studying the related technologies for constructing answering robots,and accumulating the construction experience of question answering systems.The whole project has strong theoretical significance and application value.An important test subject for the auto answering robot is the high school geography.The high school geography test questions examine the memory,comprehension,and reasoning ability of students more comprehensively.It is a kind of high-quality question answering data under the real scene.Therefore,the 863 program needs to construct a high school geography automatic question answering system with high performance for the char-acteristics of high school geography examinations.Because of the neural network model has the characteristics of text representation into dense vector,it can solve the problem by calculating the distance of the vector with-out the need of the literal matching,so the problem of incomplete template coverage in the rule system does not occur.However,when using a neural network to construct a real-world question answering system,it will face many problems.First of all,existing question answering systems based on neural networks have achieved good performance on many QA datasets,but these systems are not aimed at high school geography.Ques-tions do not consider the characteristics of geography examinations.Secondly,the high school geography examinations do not provide relevant knowledge,and the acquisition of topic-related knowledge depends on the search.Although the search introduces rel-evant knowledge to the questions,it also brings new problems.Finally,the automatic answering system in high school should show the result of the solution in a user friendly way,not just the answer.In view of the above problems,this paper explores high school geography neu-ral network based problem-solving techniques in the high school geography non-graph selection problem,and builds a neural network based automatic answering system for high school geography.Specifically,this paper can be divided into three aspects:prob-lem solving model design,problem solving strategy improvement,and problem solving system construction.The main work is as follows:1.The paper designs a neural network problem solving model for high school geogra-phy.Considering that the high school geography question itself lacks the relevant knowledge,the problem-solving framework of knowledge retrieval and knowledge encoding is adopted to design a neural network for high school geography.This au-tomatic problem solving model has a strong semantic encoder and a complex atten-tion mechanism.At the same time,the examinations and knowledge data necessary for high school geography are collected and organized.Finally,in the collected test data set,the model proposed by this paper has achieved relatively good results.2.The paper adopts a problem-solving strategy that uses the knowledge matrix to sup-port the answering process.For the inaccuracy and incompleteness of knowledge retrieval,the paper proposes the problem-solving strategy of knowledge matrix combined with neural network for the first time.The knowledge matrix is put into the model to train,and when the problem is solved,the model selects useful knowl-edge for solving the problem.Prior to building the knowledge matrix,the entity knowledge was used to enhance the sentence representation in the knowledge ma-trix.Finally,while the knowledge matrix greatly reduces the parameters required for training,the problem solving model also obtains good results on the high school geography test data set.3.The paper builds a high school geography automatic answering system based on neural network.The answering system can provide lexical analysis,knowledge re-trieval,automatic answering and other functions,and can display relevant results in a structured form to enhance the availability of the system.The answering system can be opened and used in any browser in the same local area network after being installed,without being limited by the operating platform system.The answer sys-tem constructed in the paper is the first neural network answering system for high school geography.The answering system is applied to the 863 program and has achieved certain results.The data,models,and codes used in this paper's for building the answering system are released at https://github.com/ryanlouc/gaokao_system.