A Novel Data Cleaning Framework Based on Knowledge Graph

Inreal-world applications, data cleaning has long been a challenge across both academia and industry. Unsuccessful cleaning of data may lead to inaccurate analysis and untrustworthy decision-making. This paper proposes a novel knowledge graph-based data cleaning framework. The framework performs pattern repair and inference repair on dirty data based on the obtained implicit and explicit relationships by establishing the knowledge graph and the relationship patterns in the data. The pattern repair includes both explicit and implicit relationship matching, while the inference repair includes both attribution inference repair and rule inference repair. The experimental results show that the higher the number of association relations among data tables, the greater the improvement in cleaning efficiency; moreover, the more association knowledge is contained in the knowledge graph, the more obvious the improvement of cleaning efficiency.