Research On Intelligent Question-answering System For Grain Situation Based On Knowledge Graph

With the continuous changes and developments in the domestic and international situations,the importance of grain reserve security work is increasingly prominent.However,in order to ensure effective grain reserve security work,a large amount of grain information support is essential.Therefore,a key challenge in grain reserve security work is quickly and precisely gathering grain information.A wide range of applications can be made by combining the advancement of intelligent question-answering system technology with expert domain knowledge and can help users quickly and accurately obtain the required information,improving query efficiency and accuracy.Knowledge graph-based intelligent question-answering systems can effectively solve the above problems.Knowledge graphs can organize and manage grain situation data,providing data support for intelligent question-answering systems and enabling users to quickly and accurately obtain the required information.This paper studies the intelligent question-answering system for grain reserves using knowledge graphs and proposes an improved method based on multidimensional prompt learning to solve the entity relationship extraction problem in knowledge graph construction.To provide query and question-answering services,we construct a knowledge graph-based grain situation intelligent question-answering system and establish a knowledge graph for the grains situation.The following are the primary contributions of this paper:(1)Research on entity relationship extraction models.We propose a multidimensional prompt learning entity relationship extraction(MDPL-RE)model to address the problem of insufficient feature representation in grain entity relationship extraction tasks under a few sample scenarios.The model constructs multidimensional prompt information by concatenating predefined entities,entity relationships,and prompt questions,which are then input into the PLM encoder along with the text for encoding.The global multidimensional feature extraction layer uses the SAAM structure to extract different dimensions of features,which are then fused to obtain global multidimensional features.The fused features are then combined with the hidden features to form table items,and the entity relationship extraction task is completed by filling in the table.Experimental results demonstrate that multidimensional prompt learning can obtain multidimensional semantic information contained in sentences and help improve entity relationship extraction performance.(2)Construction of grain situation knowledge graph.We first search for open-source grain situation data and preprocess it.Then we construct an ontology and a Chinese grain situation entity relationship extraction dataset for grain situation data.For knowledge extraction tasks,we adopt a rule-based method for semi-structured grain situation data,and use the BERT-Bi LSTM-CRF model for named entity recognition of grain situation for unstructured grain situation data,and use the MDPL-RE model proposed in this paper for entity relationship extraction.Finally,the knowledge collected via extraction is saved in the Neo4 j graph database to finish building the knowledge graph for the grain reserve.(3)Implementation of a knowledge graph-based grain situation intelligent question-answering system.First,we create a corpus of grain situation question sentences using pertinent rules and data augmentation approaches and complete the construction of the grain situation question sentence corpus.Then we use the BERT-Bi LSTM-CRF model to obtain grain situation question entities and the BERT-CNN model to obtain user intent,and convert them into Neo4 j query statements to search for answers in the grain situation knowledge graph constructed in this paper.The system implements the functions of each module using the Flask framework,and uses j Query’s Ajax technology to achieve data transmission between the front and back ends.Echarts.js is used to generate a visualization of the Neo4 j network relationship graph,ultimately realizing the grain situation intelligent question-answering system.