Reasearch On Relation Extraction Methods By Incorporating External Knowledge

With the advent of the big data era,data on the Internet has grown exponentially,most of which are unstructured data.Information extraction can transform unstructured data into structured data that is easy for people to understand.Among them,relation extraction is an important part of information extraction.It is also the basic work of machine translation and machine reading comprehension,which is of great significance.For some complex texts,the performance of the current relation extraction model is still not high.The main reason is that manual annotation can use external knowledge(such as common sense or experience)other than the text.However,the traditional relation extraction only uses text information.Unable to solve the problems caused by common language phenomena such as polysemy of a word,which affects the effect of relation extraction.In response to the above-mentioned problems,this paper proposes a relation extraction method that integrates external knowledge and text information,and uses external knowledge information to supplement the defects of word vectors,improve the semantic information of the vocabulary,and improve the effect of relation extraction.The work of this paper mainly includes:1.Relation extraction method based on knowledge base that integrates external knowledge.This paper proposes a knowledge base-based relation extraction method that integrates external knowledge.For the current external knowledge base such as HowNet contains rich information,the semantic common sense knowledge base is constructed using the original information in HowNet.The smallest unit of the knowledge base is the meaning It can effectively alleviate the problems caused by the polysemous phenomenon of a word.At the same time,the long short term memory(LSTM)is used to integrate the semantic common sense knowledge base into the model,and the self-attention mechanism is introduced to give each meaning to different meanings.The weight can make better use of the external knowledge in the knowledge base and further improve the word vector information.Experiments have proved that the fusion word vector can better express its semantics,and it has a certain improvement effect on relation extraction.2.Relation extraction method based on pre-training model fusion of external knowledge.At the same time,a large amount of unlabeled text information in reality can be easily obtained,and these texts contain a large amount of hidden information.If word vectors that incorporate these hidden information can be obtained,the model can obtain the text from the fused word vectors.The deep meaning of.Therefore,this paper combines the Bert model and the bidirectional gated recurrent unit(BiGRU),and uses the fused word vector obtained through the pre-training model as the input of the BiGRU model,so that the model can obtain the implicit meaning of the vocabulary and then use The attention mechanism reduces the influence of noise and further improves the effect of relation extraction.Experiments have proved that the method of fusing external knowledge based on the pre-training model can improve the relation extraction to a certain extent.