Research On Multi-hop Question Answering Method Based On Knowledge Graph Embedding

Knowledge graphs are a type of knowledge base that are widely used in many realworld scenarios,such as recommendation systems,question answering systems,and information retrieval.The knowledge graph-based question-answering method is an important research hotspot in the fields of artificial intelligence and information retrieval.Based on the length of the reasoning path on the knowledge graph,knowledge graph question answering methods can be divided into single-hop and multi-hop question answering,with the latter possibly involving time constraints or implicit information.In practical applications,on the one hand,users tend to express complex natural language questions that require strong long-path modeling capabilities.Since knowledge graphs are incomplete and missing a lot of information,this poses a greater challenge for multi-hop question answering.On the other hand,traditional knowledge graphs do not consider time information,and when facing questions with time constraints,knowledge graph-based question answering methods cannot perform well.Therefore,researchers have proposed temporal knowledge graphs.Although temporal knowledge graphs contain time information,there are few methods developed specifically for time-based knowledge graph question answering.Most existing temporal knowledge graph question answering methods focus on semantic or time-level matching,lacking the ability to reason about time constraints.To address these issues,this study mainly investigates the following two aspects:(1)A multi-hop question answering model based on relation paths and knowledge graph embeddings is proposed.The use of knowledge graph embeddings solves the problem of link missing caused by sparse knowledge graph,while the rich semantics between entities in the knowledge graph improve the accuracy of the model’s question answering.The question and relation are also represented in an enhanced manner.Specifically,a semantic extraction module for complex questions and a relation detection module are proposed.The semantic extraction module uses self-attention mechanisms to enhance the representation of questions,allowing for more accurate extraction of the multiple semantics of complex questions.The relation detection module extracts the relation paths between the topic entity and candidate entities,and finally matches them semantically with the question.The proposed model is tested on the complete Meta QA dataset and the incomplete Meta QA dataset,and the experiments show that the proposed method has high accuracy.(2)A multi-hop question answering model based on temporal knowledge graph embeddings is proposed,which consists of four modules: a knowledge graph embedding aggregation module,a question processing module,a path inference module,and an answer prediction module.The knowledge graph embedding aggregation module enhances the semantic representation of entities and relations,first using graph attention networks to enhance entity representation by incorporating current node information and neighbor node information,enriching the semantic relationships between entities and increasing neighbor information.Then,a neural network is used to aggregate time information into relation information,enriching the time-evolving information in relations.The question processing module obtains time constraint information and time constraint relations and obtains a feature representation of the question that combines context information,time information,and entity information.The path inference module can extract the subgraph related to the question topic entity and prune it according to the time constraint to obtain the relation path that satisfies the time constraint.Finally,the answer prediction module obtains the final answer.The model is evaluated on the Cron Questions and Complex-Cron Questions datasets,and the results show that the proposed method has high accuracy.