Research And Implementation Of Relation Extraction System Based On Deep Learning

In the 21st century,with the significant increase in the amount of data in the network,big data and cloud computing,which process massive data,have become mainstream technologies.Using information extraction technology,we can extract valuable and specific factual information from these massive unstructured data,and store and use these data reasonably through various databases and algorithms to create greater value.Relation extraction is a key step in information extraction.Many traditional deep learning models have achieved excellent performance in relation extraction tasks.However,the performance of these models largely depends on a large amount of labeled training data,and obtaining a sufficient amount of correctly labeled training data requires a lot of human and material resources,which makes these models difficult to be promoted well.Therefore,researching and designing a model that can adapt to fewer training samples and correctly complete relation classification has become a current research hotspot.In addition,the problems of traditional AI systems,such as algorithm iteration speed greater than system update speed,low deployment efficiency in heterogeneous environments,and low utilization of computing resources,are becoming more and more obvious.The emergence of cloud computing technology also provides solutions to solve these problems.In addition,problems such as algorithm iteration speed is greater than system update speed,low deployment efficiency in heterogeneous environment and low utilization rate of computing resources in traditional AI systems are becoming more obvious.The emergence of cloud computing technology also provides solutions to solve these problems.Through the investigation and study of deep learning knowledge such as small sample relationship extraction technology,self-attention mechanism and contrast learning,as well as the study of cloud computing,Docker,Kubernetes and other cloud technologies,this paper conducted an in-depth study on small sample relationship native extraction tasks,and designed an algorithm service system combining with cloud native technology.The main content of this paper includes the following three aspects:(1)An external concept fusion method based on selection gate and self-attention mechanism is proposed.First,the candidate concepts are selected by introducing the concept of an external database to the entities in the sentence.Then,the similarity gate is used to determine the relevance of each candidate concept,leaving the concepts that can help classify and filtering out the unrelated concepts.Finally,the self-attention mechanism is used to integrate the concepts at character level to enrich the input semantics.(2)This paper proposed a relation extraction method for small sample based on contrastive learning.Aiming at the problem that the ability of the model to discriminate similar relations in the same batch is reduced in the case of small sample size,this paper introduces a supervised contrastive learning loss,uses relation descriptors as anchors,the same relation instances as positive samples,and different relation instances as negative samples,and optimizes the distribution of relation prototypes in the vector space to increase the discriminant ability of the model.(3)The algorithm service platform based on cloud native is realized.The platform provides algorithm uploading,algorithm management and algorithm usage functions.With Docker and Kubernetes as the support,the system decoups the algorithm from the system and the physical machine,which improves the operation and maintenance efficiency of the algorithm system and the utilization rate of physical resources.