Research On The Extraction Of Entity Relationships From Multivariate Information

Information extraction is a research hotspot in the field of Natural Language Processing(NLP),and entity relationship extraction is particularly important as its basic task.The purpose of entity relationship extraction is to strip out a structured entity from a large number of unstructured data and then judge the semantic relationships between the entities.It is usually expressed in the form of triples.This technology provides a basis for knowledge graph building,information retrieval,automatic question-and-answer and other technologies for the future.At present,most of the research directions of entity relationship extraction are to represent it with a single word vector or combine different features to process it.The results processed in multiple methods have their own advantages and disadvantages in relational judgment in the classification of different relationship types.Therefore,this paper proposes the core innovation point that multiple data processing methods are adopted to process the experimental data,to input them separately to the next layer,to count their advantages and disadvantages and to assign weights to them more rationally.The research of relationship extraction was also developed step by step,which went through knowledge engineering,traditional machine learning and deep learning.At present,deep learning methods have become the mainstream today.The reason is that relative to those traditional entity relationship extraction methods requiring professional design characteristics,deep learning methods effectively prevent spending a great amount of manpower and time as traditional methods through automatic learning and model statistics rules,and reduce the spread of error.Therefore,this experiment was carried out based on deep learning.However,in the deep learning model,it is also found that a reasonable combination of the convolutional neural network and the bidirectional LSTM model can also improve the overall performance of the experiment well.In order to highlight the importance of the entity in the sentence and better explore the relationships between words,the attention mechanism is incorporated into the neural network model.In terms of the above problems,the corresponding solutions proposed in this experiment have the following three main innovation points;(1)In this experiment,a multi-model fusion strategy was proposed,in which the syntactic structure analysis,the shortest dependency path and the results processed by the How Net are inputted into the neural network layer separately,and the weight distribution of different models for different relationship recognition is optimized based on the sensitivity of each method in the corresponding relationship classification.(2)On the basis of the original model of BI-LSTM,the convolutional neural network(CNN)is added,so that it is possible to better capture the partial key information of sentences,so as to improve the performance of the experiment.In order to maintain data integrity,every input is combined with the original data.The purpose is to maintain the integrity of the data information and input it into the BI-LSTM layer in a better state,so that the advantages of both models are well expressed.(3)In order to better study the dependency relationships between words and capture the internal semantic information of sentences,the attention mechanism is incorporated into the original model.In the attention layer,in order to improve the classification effect in the long complex sentence,all the hidden states of the data inputted into the attention layer are extracted.In order to prove the effectiveness of this method,the contrast experiment was conducted.Finally,the results prove that it is better to extract all the hidden states than to extract a single hidden state.Compared with the experimental results,after the integration of the multivariate information neural network model,the F value was 79.38%,which was significantly improved compared with the initial experiment.After the attention mechanism model was incorporated,the F value increased by 2.62%.The experimental data proved that the experimental method was feasible.