Design Of Question Answering System Based On Knowledge Graph Reasoning

As a characteristic advantageous industry in Ningxia,the wolfberry industry occupies a very important position in the rural economy of all districts in Ningxia.In addition to the impact of market prices,there are two factors that have the greatest impact on the production of wolfberry in Ningxia:one is climate,and the other is disease and insect pests.In order to learn more about the pests and diseases of wolfberry,farmers need to use the Internet to learn this knowledge.In the past,knowledge was retrieved through search engines.With the rise of the knowledge graph,the knowledge graph can be used to obtain the knowledge that users really need more quickly and effectively.It is the knowledge graph that organically organizes the information according to the knowledge relationship.,So that the information has an inherent knowledge structure,which not only can effectively solve the problem of information overload,but also brings the possibility of providing users with intelligent services.Therefore,the question answering system based on the knowledge map of wolfberry diseases and insect pests will certainly provide users with more accurate and intelligent question answering services.This paper expands the research and design of the question answering system based on the knowledge graph of wolfberry diseases and insect pests.First of all,this paper believes that the core question in the question answering system is to clearly understand the user's question.The three-tuple formal representation of entity recognition and question sentences;secondly,how to effectively find the answer in the knowledge graph is one of the key questions of the intelligent question answering system.This paper uses the vectorized representation model of the knowledge graph to vector the knowledge graph;at the same time,the analytic question is also expressed by vectorization,and finally the corresponding answer is determined for the user in the knowledge graph through the similarity comparison.The main research content has two aspects:(1)Based on knowledge graph representation learning to realize question analysis.Question parsing is the primary task of the question answering system.Its main purpose is to decompose the semantic information in the user's question through syntactic analysis and find the key entities in the question.First,use the entity naming recognition model BiLSTM-CNN-CRF to perform entity recognition on the question sentence,and clarify the core part of the question sentence in the knowledge graph.Secondly,use tools such as part-of-speech tagging and syntactic analysis to decompose the question sentence,because the question sentence can only determine the core word of the question sentence and the predicate relationship of the question,and the missing object part is the answer that the user needs,so,Question analysis extracts the subject-predicate structure of the question and transforms it into a triple form.The TransH model can not only introduce the rich semantic information in the knowledge graph,but also learn the correct representation of entities and relationships.TransH the vectorization method vectorizes the triple structure of the question sentence,so that the TransH model is used to predict the correctness of the question triplet in the knowledge graph.(2)Realize answer generation based on knowledge graph representation.In this paper,the research on knowledge representation is carried out on the knowledge map of wolfberry diseases and insect pests,and the TransH model is used for representation learning of the structural characteristics of relationships and entities.First of all,in this article,the entities parsed by question sentences are used to link entities in the knowledge graph using SPARQL,and then all the triples of the knowledge subgraph with this entity as the core are taken out as alternative answers,due to alternative answers and questions They are all expressed in a vectorized form.The method of selecting a similarity measure calculates the similarity between the triples of the question sentence and the candidate answers and ranks them.The candidate answer with the highest ranking is taken as the answer and returned to the user.