Text Entities And Their Relationship Mining Based On Feature Fusion

With the development of the Internet,a large number of texts are filling people's lives.Information extraction can help people to obtain important information quickly.The research on information extraction has become more and more important.Entity recognition and relation recognition are the basic technologies for information extraction.The extraction of entities and relations is also an essential key technology in the application fields of constructing knowledge graphs,realizing semantic search,and establishing intelligent question answering systems,and has extremely important research value.Therefore,this thesis studies the joint extraction of entities and relations from two aspects: parameter sharing and joint decoding,and constructs a prototype system for joint extraction of entities and relations in the field of tourism.The details are as follows:1)A joint extraction method of shared parameter entities and relations based on multi-feature fusion is proposed.Most of the previous entity and relation joint extraction models only considered the Bert pre-training model to obtain the bidirectional features of words,but did not consider the sentence features and character features.Therefore,a model of joint extraction of shared parameter entities and relations that fuse characters,sentences and word features is proposed.Our method will extract the character features by multi-layer convolution neural network,use convolution neural network with different sizes of convolution kernels to extract the sentence features,and use Bert pre-trained language model to extract word features.This can better capture various features in sentences,and improve the robustness of the model by using negative sampling in training.2)Based on the joint decoding of entity attention mechanism,a joint extraction method of entity and relation is proposed.In the previous joint extraction methods of entities and relations,the joint model did not make good use of the information between entities and relations,so a joint extraction model of entities and relations based on entity attention mechanism is proposed.Our model will obtain the semantic features of words in sentences by bidirectional long and short memory network(Bi-LSTM),and use the entity attention mechanism to dynamically extract the important word features and category features of the entity for relation extraction.Thus,the interaction between entity and relation joint extraction is improved.3)A prototype system for joint extraction of entities and relations in the tourism field is constructed.Previously,entity recognition and relation recognition in the field of tourism used traditional supervised methods,relying on a large number of manually annotated data,which was time-consuming,labor-intensive,and expensive.Therefore,a joint extraction model of entities and relations based on Bi-LSTM and CRF is used.The model uses Bi-LSTM methods to learn the features of words with minimal manual dependence,and uses CRF to optimize the sequence annotation results in entity recognition.The model uses Bi-LSTM to extract the features of words,the entity features after entity recognition and the distance features between entity pairs,so as to complete the relation recognition in model and realize joint extraction of entities and relations in the tourism field.Finally,the model is used to complete the prototype system for the joint extraction of entities and relations in tourism field.Experimental results verify that the methods proposed in this paper are effective.F1 value of entities and relations in entity relation extraction based on multi-feature fusion are 86.62 and 72.29,respectively.The model based on entity attention mechanism can also effectively identify entities and relations.The model based on Bi-LSTM and CRF used in the tourism field can also complete the extraction of entities and relations.And a prototype system for joint extraction of entities and relations in tourism field is successfully constructed.