This paper explores a novel procedure for creating Knowledge Graphs (KGs) from unstructured text to improve the management of news story data. The method leverages iterative few-shot Large Language Model (LLM) prompting and unsupervised clustering models, combined with state-of-the-art text embedding techniques, to disambiguate entities and predicates. It also incorporates node and edge attributes to enhance graph analytics. The approach addresses multi-language text handling and provides KGs in various languages. By translating natural language queries into graph queries, this method improves user interactions and information retrieval. An evaluation protocol is introduced to assess the effectiveness of this procedure, aiming to enrich news content by identifying and contextualizing relationships between entities, thus enhancing data insights.
Graphs can enhance understanding of data by gathering different sources of information into a web of knowledge with context Machine Learning models can identify entities and relationships in unstructured data to enhance knowledge graphs The enhanced knowledge graphs can help with news story research and creation LLM techniques can be used to more effectively query a knowledge graph