Research And Application Of Key Technologies Of Knowledge Graph-based Question Answering System Using Deep Learning

Knowledge graph provides a hierarchical triple knowledge query for question answering systems,and is currently a research field that is widely concerned.However,in the Chinese question answering system based on the knowledge graph,many existing methods can only deal with some Chinese questions with simple structure,that is,the query can only be linked to the knowledge base through a single triplet to obtain results,which cannot be a good solution to deal with the complex questions that contain parallel multiple entities and multi-layer inference.Therefore,how to design different solutions for different types of Chinese questions has become a key problem to be solved in the field of knowledge graph question answering.In response to this problem,this paper conducts an inductive analysis of the diversity of Chinese questions and the search methods of query paths,and uses deep learning and pre-trained language models to study the key technologies of knowledge graph question answering.The specific work of the paper is as follows:1.Research on multi-label Chinese question classification method based on mixed modelAiming at the problem of the diversity of questions in Chinese question answering systems,a multi-label labeling strategy is proposed to classify Chinese questions according to different structures.According to the SPARQL statement of the question sentence,the questions are divided into simple direct result-oriented questions and complex indirect result-oriented questions,so as to achieve different solutions for different types of questions in practical application scenarios.In the question processing stage,in view of the low accuracy of boundary recognition in existing named entity recognition methods,a hybrid model-based named entity recognition method is designed.2.Research on information extraction and Answer retrieval methods based on knowledge graphAiming at the problem that an entity mentioned in the Chinese question may correspond to multiple entities in the knowledge graph,an entity link method using multi feature screening is proposed to select the candidate entity with the highest priority for entity link.The semantic similarity algorithm is used to establish the relationship extraction classification model to obtain the candidate relationship closest to the entity.In the stage of obtaining answers,different solutions are taken according to different types of questions to get the answer path of questions.3.Development of Chinese Knowledge Graph Question Answering Prototype SystemIn order to verify the effectiveness of the research method,a Chinese knowledge graph question answering prototype system is constructed based on the above research content.Firstly,the system requirements are analyzed and the overall system architecture is designed.Based on the Flask framework,the system is developed to realize friendly questioning and answering for complex questions in non-reasoning questions raised by users.The results of Question Answering and related knowledge recommendation are presented in the form of visualization.The function of searching the answer according to the question type in the knowledge graph is realized.