Research on Port Knowledge Graph for Business Analysis

To cope with the challenges of multi-source heterogeneous data integration and complex text semantic understanding in port intelligent decision-making, this paper proposes a word boundary perception model based on multi-task learning ( MULTI-WBPM ), which aims to improve the construction efficiency and accuracy of the port business knowledge graph. The research focuses on the key technologies to improve port domain entity recognition and relationship extraction. By combining static embedding ( BERT general semantics ) and dynamic embedding ( domain character features ), combined with bidirectional GRU to enhance boundary perception, the recognition performance of long entities and professional terms is significantly optimized. In the relationship modeling stage, the joint analysis method of rules and syntax tree is designed to effectively solve the problem of semantic ambiguity under complex sentence patterns. In terms of knowledge fusion, an entity alignment strategy based on the similarity between Sorensen-Dice coefficient and the BERT vector is proposed, and knowledge retrieval is accelerated by combining graph database index. The experimental results show that the entity recognition F1 value of MULTI-WBPM on the port business data set reaches 95.50 %, which is better than the baseline model, and the decoding efficiency is improved by 25%. The constructed port knowledge graph integrates multi-dimensional data such as logistics, equipment, and security to support dynamic resource scheduling and risk early warning. The research verifies the effectiveness of multi-task learning and knowledge graph technology in port intelligent decision-making, and provides a scalable technical framework for port operation optimization and green sustainable development.