The Construction And Application Of Domain Knowledge Graph On The Science And Technology Service

With the development of science and technology service industry,there is an urgent need to effectively organize the multi-source heterogeneous science and technology resources to meet the personalized needs of different industrial fields for science and technology services.The science and technology resources required by different industrial fields are provided according to the field perspective,including science and technology patent resources,science and technology paper resources,science and technology consulting resources,science and technology project resources,and instrumentation resources,etc.The knowledge graph adopts the form of triad to describe the entities in the text and the relationship between them,and construct the semantic association between the knowledge.By constructing a domain knowledge graph oriented to industrial industry science and technology services,science and technology resources can be effectively organized and managed,and the science and technology resources in a certain industrial field can be accurately and quickly retrieved.However,science and technology service involve many industrial fields,which are highly specialized and changeable,so how to construct the domain knowledge graph for science and technology services is the main research content of this paper.The main contributions of this paper are:(1)To address the problems of complexity of technical terms and the difficulty in identifying entity boundaries in scientific and technological resources,and it is impossible to effectively extract the long-distance semantic features of texts,This paper proposes an entity extraction method in the field of scientific and technological services based on graph convolution network(GCN).Firstly,the character vectors obtained by BERT layer are expanded and supplemented by adding additional part-of-speech features to obtain semantic information between characters;Then,on the basis of bi-directional long short-term memory(BiLSTM),it is integrated into graph convolution network,which is used to mine the structural information of characters and the relationship between characters.In addition,the feature representation extracted by BiLSTM is spliced and fused with the dependency matrix between characters to fully obtain the global features of the text.Finally,the feature vectors obtained by GCN layer are sent to CRF model for sequence decoding,and the global optimal sequence is selected,which is the result of entity recognition.The experimental results show that this method is superior to the traditional entity extraction method and can effectively improve the accuracy of entity extraction of scientific and technological resources.(2)In order to solve the problem that the relationship between scientific and technological resources entities is complex and lacks a large number of labeled data,which makes the prediction of relationship labels wrong and produces a lot of noise,this paper proposes an relation extraction method in the field of scientific and technological services based on comparative learning.Firstly,the BERT model is fine-tuned based on contrastive learning,and the pre-training language model SciBERT suitable for scientific and technological services in industrial fields is obtained to capture the contextual semantic features of text sentences.Then graph attention network is used to obtain the structural features of sentences,and greater attention weight is set for nodes with important semantic expression.Finally,the feature fusion method based on location attention is used to fuse the contextual semantic features and sentence structure features of the sentence to obtain the text sentence representation of scientific and technological resources,and the Softmax classifier is used to decode and obtain the relationship prediction label.Experiments are carried out on the data set of scientific and technological resources in the industrial field,and the experimental results show that this method can effectively improve the effect of relationship extraction.(3)Knowledge graph construction and application in the field of science and technology services.Based on large-scale actual science and technology resources such as patents and papers,the extracted knowledge is stored in the Neo4j database.Further,a prototype system for the knowledge graph construction and visual analysis in science and technology service fields is developed and designed,which is convenient for the query and utilization of the knowledge graph and the reorganization of scientific and technological resources for industrial sectors.