Deep Learning-Based Joint Extraction Model of Entity Relationships for Cloud Operations Knowledge Graph

Knowledge mapping can effectively integrate effective information in massive data, laying the foundation for semantic intelligent search and knowledge interconnection. At present, with the digital transformation and upgrading of the grid cloud platform, the cloud platform operation and maintenance process exists the problem of a weak correlation between information and low efficiency of decision generation, and a large number of operation and maintenance data resources of the cloud platform have not been integrated and utilized. In this paper, we study the joint extraction of entity relationships in the knowledge graph, design the structure diagram of the cloud O&M knowledge graph, propose a deep learning-based Seq2Seq joint extraction model, and the encoding layer obtains the contextual semantic representations through BERT-WWM and Bi-directional Long Short-Term Memory (BiLSTM) models and incorporates the multi-attention mechanism to highlight important words in the text. The decoding layer outputs sequence annotations through a pointer network to obtain head entities, relations, and tail entities. Using the self-annotated dataset as the corpus, the quality of entity and relation extraction is evaluated by modeling with the TensorFlow framework. The results show that the precision, recall, and F1 value of the model are high, which validates the effectiveness of the joint extraction model.