Research And Implementation Of Entity Relation Extraction In Knowledge Graph Based On GCN Technology

With the development of technology,there are more and more mobile terminals and smart devices,and the interconnection of everything has become possible.The explosive data growth caused by interconnection has made the amount of data in the virtual space more and more huge.In the field of machinery industry,all kinds of raw materials,components,processing technology,technical performance,national standards and other data are more complex and diverse.How to assist workers in related fields to accurately and efficiently find target data and related data from these data? Relational data is even more important,and the requirements for data search and other related technologies are also increasing day by day,and knowledge graph technology emerges at the historic moment.The knowledge graph can effectively process a large amount of data existing in the real world,extract entities and the relationships contained therein,and store them in the database as knowledge,thereby facilitating various operations on these data.And the logic in the knowledge map is very in line with people's way of thinking.For a specific task,as long as there is a need for relationship analysis,the knowledge map is likely to come in handy.As one of the cornerstones of the knowledge graph construction task,the relation extraction task is an indispensable step.Accurate and perfect relation extraction effect is critical to improving the performance of the knowledge graph and the system tools based on the knowledge graph.Among the existing relationship extraction technologies,methods based on supervised learning can use high-quality labeled data to learn a relationship extraction model.However,the acquisition of high-quality labeled data is difficult and requires great manpower and material resources.In the field of machinery industry,only one type of workpiece can have multiple parameters,such as its raw materials,uses,technical indicators,processing technology,etc.,and with the development of time,the variety of workpieces is also under development.The number of such workpieces in the machinery industry is huge,and each individual workpiece is marked,and the workload is extremely huge.In relation extraction methods based on unsupervised learning,deep learning techniques are used to learn relation extraction models from a large amount of unlabeled data.However,in actual use,the named entity recognizer needs to be used,and the named recognizer itself has a certain error,which will cause the expected performance to decrease.Due to the particularity of the industry,the mechanical industry has higher requirements for the comprehensiveness and accuracy of data.At the same time,there are mutual influences between the relationships.There are some entity pairs that have more than one relationship,and this kind of interaction is very common in the knowledge base.For example,in the field of machinery industry,between "alloy steel" and "milling cutter",alloy steel can be used to manufacture milling cutters,and milling cutters can also be used to process alloy steel.However,the existing methods rarely deal with this problem.deal with.In response to the above problems,this article proposes the following research content.(1)In order to ensure the accuracy and comprehensiveness of the relationship extraction results,an end-to-end relationship extraction method based on LSTM and GCN technology is proposed.The end-to-end model can complete entity recognition tasks and relationship extraction tasks at the same time,thereby avoiding error propagation problems.In addition,the LSTM+GCN method can extract as many features as possible from the input corpus.When used in relation classification tasks,it can obtain as rich and accurate relation classification results as possible,providing a reliable basis for subsequent tasks.(2)In order to deal with the situation of overlapping entities,and to consider the interaction between relationships and the interaction between relationships and entities,the relationship weighted graph and GCN are used to improve the model.GCN is used to extract the comprehensive word features contained in the relationship weighted graph,and through the assistance of the entity loss function relationship loss function,so as to achieve more accurate and more fully considered relationship extraction results.A large number of traditional methods have not dealt with the relationship between entity overlap and mutual influence,and the accuracy rate will be negatively affected.The improved model can handle a lot of entity overlapping relationships that exist in the mechanical industry field,so as to achieve a better relationship extraction effect.(3)Based on the above research content,this paper implements a joint extraction system of entity relationships for the mechanical industry.At present,the mechanical industry still relies more on traditional technical manuals and other tools.These tools are slow to update and inefficient.After completing a set of data queries,other tools need to be used to query another set of data,which is not easy to find the relationship between various data indicators..Through the development of this system,it can assist workers in the relevant machinery industry to efficiently and accurately complete the work of component screening,process formulation,raw material procurement and production maintenance.