Knowledge Graph Construction And Modeling Analysis Of Grain Postpartum Loss

In China,the resources are extremely scarce and the relationship between grain supply and demand is tight for a long time.Therefore,reducing post-harvest losses has great potential and important strategic significance.With the development of technologies including Food Internet of things and so on,the data on post-harvest loss of grain presents many problems such as rapid growth,numerous sources,complex structure,and difficulties in access and management.In particular,the problems of numerous sources and complex structures directly restrict the efficient retrieval and correlation analysis of food loss information.Therefore,it is very important to generate a relationship model between post-harvest loss variables,construct a post-harvest loss data system,and establish a post-harvest loss classification model,which can help to reduce grain loss.In view of the above problems,this thesis investigates the modeling and analysis of grain postpartum loss data after building a knowledge graph and based on its entity attribute information.The specific work is as follows:Firstly,based on the scrapy framework,relevant data of the statistical bureau,the grain bureau and other websites is crawled,which is cleaned,labeled and integrated into a corpus.Further processing of corpus includes word segmentation and part-of-speech tagging.At the same time,the data is filtered and classified based on heuristic rules and k nearest neighbor algorithm,where 14 entity categories are obtained including post-harvest loss,botanical nouns and so on.The results of entity extraction experiments show that the weighted classification accuracy rate of k nearest neighbor algorithm is 5.3% higher than that before weighting,which verifies the effectiveness of this method in entity identification.Secondly,the problem is extracted for the relationship between entities,and its task is abstracted into a classification problem.The lexical information and syntactic information are obtained by syntactic analysis,which generates the syntactic parse tree of sentence.Then,the semantic information is obtained through the specific structure of the sentence.Moreover,the entity relationship is classified on the basic of the CNN algorithm model and the PCNN algorithm model.The classification is mainly divided into seven inter-entity relationships,such as the nature and the superior classification.Finally,the relationship between the entity and the entity is used to construct a map of the post-harvest loss knowledge and visualize it.The experimental results show that the classification accuracy of the PCNN model is 7.6% higher than that of the CNN model.It also shows that the performance of the PCNN model is better than that of the CNN model for the entity relationship extraction,especially for the segmentation operation of the sentence,the text feature can be better represented.Thirdly,on the basis of the post-harvest loss knowledge graph,relevant loss factors,such as harvest time,pest degree,etc.are obtained,and relevant loss data are obtained based on the loss factor.Combined with k nearest neighbor,logistic regression,decision tree,XGBoost and other algorithms,a multi-model fusion classification method for post-harvest loss of grain is proposed.This method mainly divides the degree of food loss into four categories: “very few”,“general”,“serious” and “extremely serious”.The experimental results show that the multi-model fusion classification method proposed in this thesis is relatively ideal for the classification of post-harvest loss of grain since three evaluation indicators(Recall: 94.0%,Precision: 94.0%,F1-Score: 93.2%)were superior to the traditional classification model.In addition,this thesis also designs and implements a knowledge graph system for post-harvest loss,which enables entity identification,relationship query,and food knowledge overview.