Research On The Construction Methods Of Knowledge Graph Of Dangerous Goods

Unconventional emergencies are different from other events that cause harm to public safety,mainly in the areas of suddenness,unpredictability,uncontrollability,evolutionary change,etc.Which will often bring huge bosses to personnel and social property,seriously affect our living and endanger public safety.Dangerous goods are an important source of unconventional emergencies,and the management of dangerous goods storage,transportation and other links often lead to accidents and even catastrophes due to poor handling.If there is a tool for quick access to relevant knowledge,the management and training cycle will be greatly reduced.In view of this realistic demand,the paper proposes the use of knowledge graph technology to construct a special knowledge graph in the field of dangerous goods.At present,no knowledge graph of dangerous goods in Chinese has been found in the survey.Therefore,the article intends to take this as the goal,focusing on combing the framework of the construction of knowledge graph of dangerous goods,using two broad categories of methods to construct knowledge graph,compared with the results of the artificial.And the results show that both methods show better performance in entity extraction,but the word vector similarity based on deep neural network under the corresponding optimal parameters is superior to the Co-occurrence model both in entity relations extraction.It is validated that the method proposed in this paper is effective and feasible,and through comparison and analysis,find a better method and corresponding model parameters.The article mainly including the following work:(1)Constructed the technical framework of knowledge graph of dangerous goodsThe nature of dangerous goods is very complex and the relationship between substances is numerous,so a variety of related technologies will be needed in the construction process.Here we carried on the whole framework design of technology involved and to satisfy the construction of knowledge graph of dangerous goods.(2)Establish two major categories of dangerous goods entities and relationship extractionOn the basis of entity extraction,we focus on the extraction of entity relations,and adopt two ways: based on co-occurrence and word vectors to extract the relationship of entity.Build eight kinds of entity relations extraction based on Co-occurrence model,and with the deep neural network to train the corpus of dangerous goods to product word vector,then build a model to extract the relationship of entity.(3)Comparative assessment of construction methods of dangerous goods knowledge graphFor the two model,Co-occurrence model and Word Vector Similarity Model,to extract entity relationship,we compared and analysis them based on matrix,then compared with the expert knowledge model.In view of the three dimensions,different divisions,different thresholds and different contrast indicators to detailed comparative analysis of two models,and after the two model compared with the model of based on the expert knowledge,then determined the best construction method and model parameters.