Research On The Deep Learning Model Integrating Knowledge Graph

Driven by the big data,artificial intelligence technology has made great progress,especially in the fields of deep learning and knowledge engineering.Compared with machine learning,the development of deep learning has made a qualitative leap on the basis of large-scale labeled data.Through training data,deep neural networks can obtain ideal hierarchical feature representations.However,with the consumption of data advantages,deep learning models have gradually demonstrates the over-reliance on labeled data and lack of expert experience and knowledge.At the same time,as the scale of the knowledge graph continues to increase,the human prior knowledge and rich structured information contained in it do not play a role in deep learning.Therefore,in the current development of deep learning,the effective combination of knowledge graph and deep learning is an important method to improve the performance of deep learning models.The main work of this article is as follows:(1)To solve the problem that the existing knowledge graph embedding methods cannot effectively retain entities and ignore the relationship correlation in the complex relationship graph,this paper studies and improves the knowledge graph embedding method based on the translation model,by analyzing the relationship between the intrinsic correlation,grouped by correlation model,vector space of improvement ideas are given,and determine the objective function and training algorithm of the improved model.By using real data to comprehensively evaluate the vectorized representation of the knowledge graph to provide high-quality continuous knowledge vectors for deep learning models.(2)Aiming at the problems of large-scale data labeling of traditional deep learning models and difficulty in using prior knowledge,this paper designs and builds a model KGDe L integrating knowledge graph and convolutional neural network(CNN)on basis of knowledge graph vectorization.The structured knowledge in the domain knowledge graph is obtained through entity link disambiguation and knowledge graph embedding,and the feature word vectors and corresponding knowledge entity vectors in the original corpus text are used as multi-channel input of CNN.In the process of convolution,different types of text features are represented from the two levels of semantics and knowledge.Meanwhile,the overall framework of the KGDe L model,the network structure of each part and the fusion method are analyzed in detail,and the most suitable preprocessing method is selected for the input vector of model to determine the loss function and training method of the model.(3)Apply the fusion model to the field of disease diagnosis.This paper uses semantically rich and standardized Chinese medical knowledge graph and disease description text data to train the model.Through a series of parameter selection experiments,the parameters that optimize the performance of the model are determined;at the same time,design multiple sets of comparative experiments for the diagnosis of one or more diseases to verify the effectiveness of the model proposed in this paper,and by comparing with other disease diagnosis methods,it is shown that this method of joint training of knowledge and data is more suitable for the preliminary diagnosis of diseases based on the description of the disease.