The paper presents a study on using language models to automate the construction of executable Knowledge Graph (KG) for compliance. The paper focuses on Abu Dhabi Global Market regulations and taxonomy, involves manual tagging a portion of the regulations, training BERT-based models, which are then applied to the rest of the corpus. Coreference resolution and syntax analysis were used to parse the relationships between the tagged entities and to form KG stored in a Neo4j database. The paper states that the use of machine learning models released by regulators to automate the interpretation of rules is a vital step towards compliance automation, demonstrates the concept querying with Cypher, and states that the produced sub-graphs combined with Graph Neural Networks (GNN) will achieve expandability in judgment automation systems. The graph is open sourced on GitHub to provide structured data for future advancements in the field.

翻译：本文针对利用语言模型自动化构建用于合规的可执行知识图谱进行了研究。研究聚焦于阿布扎比全球市场的法规与分类体系，涉及对部分法规进行人工标注，训练基于BERT的模型，并将其应用于其余语料。采用共指消解与句法分析解析已标注实体间的关系，并构建存储于Neo4j数据库中的知识图谱。文章指出，利用监管机构发布的机器学习模型自动化解释规则是实现合规自动化的关键步骤，通过Cypher查询语言验证了该概念，并表明所生成的子图结合图神经网络（GNN）将实现裁决自动化系统的可扩展性。该图谱已在GitHub上开源，旨在为领域未来发展提供结构化数据。