Knowledge Graph-Based a Purchase Order Knowledge Retrieval Model (POKREM) for Steel Plants

Due to the aging of steel plants, repetitive equipment replacements and maintenance are rising. As equipment purchases increase, the engineer's workload of reviewing purchase order (PO) documents also grows proportionally. This study aims to develop a knowledge retrieval model for documents using knowledge graph (KG) technology. The target documents are PO documents related to equipment purchase contracts for steel plants. To develop the model, we first defined the data hierarchy of the steel plant. The authors distinguished relationships between nodes and created a graph database by writing triples. Ultimately, we used Neo4j to develop a purchase order knowledge retrieval model (POKREM). POKREM is a model that utilizes semantic search and KG technology. When the engineer enters a question about PO data, the model automatically extracts relevant information and provides it in a visualized form. The model's test results showed an accuracy of 99.7%, a precision of 91.7%, a recall of 100%, and an F1 score of 95.7
^%
. POKREM is expected to improve work efficiency by shortening equipment PO review time. However, there is a limitation: when humans ask the model questions, it can only be done through queries rather than natural language. Further research can be expanded by adding automatic recognition features and natural language queries to PO documents. Auto-recognition of PO documents and natural language queries can expand future research.