Joint Extraction Model Of Entity Relationship In Capsule Network Based On Part Of Speech Weighting

With the development of science and technology,science and technology policies are increasing year by year.How to efficiently extract the required information from science and technology policies has become an urgent problem to be solved.Relying on the "Hebei Science and Technology Innovation Big Data Public Service Platform",the research and implementation of the technology policy entity relationship extraction model.This paper builds a joint extraction model of entity relationships based on capsule networks to extract technology entity relationships from technology policies;In order to further improve the operating efficiency of the model,a joint extraction model of capsule network entity relationships combined with part-of-speech weighting was constructed.The main research contents are as follows:(1)Joint extraction model of entity relationship based on capsule networkAccording to the characteristics of science and technology policy,this paper proposes a joint entity relationship extraction model based on capsule network.For the problem with strong domain of text content,word2 vec is used to train targeted word embedding vectors to improve the accuracy of text data expression;for the case where the importance of word vectors is inconsistent,self attention is used to weight word vectors;for the case where the structure of convolutional neural network is single and the pooling layer loses more information,capsule network is used to enrich knots At the construction level,text information is preserved by capsule,and dynamic routing is used to improve the classification effect.The feasibility of the proposed model is proved by experiments,and the single variable method is used to test the influence of several important parameters on the experiment.(2)Combined part-of-speech weighted joint relationship extraction model for capsule network entity relationsIn this paper,we build a joint extraction model of entity relationship of capsule network based on two-way gating cycle unit generation,and prove that the macro average F1 value based on BLSTM model is higher than that based on BGRU model,and the running time based on BGRU model is shorter than that based on BLSTM model;In view of the incomplete semantic expression of word vectors and considering that the capsule network contains certain position information,two kinds of part of speech weighting models based on self-attention are constructed by adding part of speech features to enrich text information,and the validity of the later combination model is verified by experiments.For the distribution of the weight of word features and part of speech features,an artificial fish swarm algorithm is proposed The method of iterative optimization optimizes two feature weights to improve the classification effect.The effectiveness of the two optimization schemes is proved by experiments.