Construction and application of knowledge graph in mechanical fields based on deep learning

Aiming at the problem that it is difficult to reuse the design information stored in a decentralized form within the enterprise, the design model library, and the text resource library are integrated to build a knowledge graph in the field of mechanical design. Obtain the required domain knowledge from historical model libraries and documents through computer vision and other technologies: for model library data, clustering algorithms are used to reduce map node redundancy; A new machine ALBERT pre-trained model is proposed to design parameters for universal ALBERT, which is used as the semantic coding layer of text, Bi-LSTM as the label prediction layer, CRF as the overall label optimization layer, and the entity recognition model is built. For the relationship extraction model, the character encoding layer is shared, the CNN layer is added to classify the text relationships, and the unstructured information is converted into a structured triple, which is stored in the graph database after entity alignment. A computer-aided design visualization system is established based on the completed map, which aims to provide various functions such as entity query and knowledge quizzes to improve product design efficiency.