Hybrid Intelligence for Knowledge Graph Construction

知识图（KG）[11]已经成为表示知识的最强大的机制之一，并已被证明在许多应用中发挥重要作用，包括聊天机器人开发，数据集成和语义搜索[17]。然而，大多数数据源通常仍然通过异构的非图形数据结构，方案和格式表示[7]。将这些视图转换为KG视图仍然需要KG工程师和领域专家在完成管道中的每项任务时付出相当大的努力，证明这是一个耗时且通常复杂的过程[1]。虽然在支持KG工程师工作方面取得了进展，但“现有技术提供了[.]解决方案，每一个都涵盖了KG建设的一个特定方面，但自动编排[...]整个建设过程仍然是一个挑战”[12]。这一挑战激发了我们的研究。因此，我们的目标是（i）在数据集成过程和自动化的背景下丰富对KG施工的理论理解，（ii）将人工智能的应用领域扩展到KG施工任务到管道的编排，以及（iii）通过为KG施工管道设计提供新颖的方法来支持从业者社区，生成和自动化。我们建议首先研究和建模KG施工管道，以发现和表示一般特征和操作。然后，基于KG工程师的行为，将其理解为面向目标的代理，我们将提出一个混合智能problem-solvingagent的设计，作为这些数据集成管道的自动编排的可能解决方案，专注于结构化和半结构化的源。

Knowledge Graphs (KG) [11] have risen to be one the most powerful mechanismsto represent knowledge and have been shown to play an important role for manyapplications, including chatbot development, data integration, and semanticsearch [17].However, most data sources are generally still represented via heterogeneousnon-graph data structures, schemes, and formats [7]. The conversion of thesesources into KG views still necessitates a considerable effort from both KGengineers and domain experts when completing each of the tasks in the pipeline,proving this to be a time consuming and often complicated process [1].While advances have been made in terms of supporting KG engineers in theirwork, "existing techniques offer [...] solutions, each covering a specific aspect ofKG construction, but automatically orchestrating [...] the whole constructionprocess remains the challenge" [12]. This challenge motivates our research.Therefore, our objectives are (i) enriching the theoretical understanding ofKG construction in the context of data integration processes and automation,(ii) extending the application domain of AI to the orchestration of KG construction tasks into pipelines and (iii) supporting the community of practitioners bysupplying them with novel methods for KG construction pipeline design, generation and automation.We propose to undertake this by first researching and modelling KG construction pipelines to discover and represent general features and operations.Then, basing ourselves on the actions of the KG engineer understood as a goal oriented agent, we will propose the design of a hybrid intelligent problem-solvingagent as a possible solution to the automatic orchestration of these data integration pipelines, focusing on structured and semi-structured sources.