Identifying relevant tables in databases to build event logs is typically a manual, error-prone task in process mining. This paper introduces TabMine, a semi-automated algorithm that identifies these tables by leveraging both the network structure of tables and their natural language descriptions. By integrating process-related business documents with table metadata, TabMine employs machine learning techniques, specifically community detection and natural language processing, to align table communities with the corresponding documents. This enables analysts to build event logs for process mining from a targeted list of tables without prior knowledge of the specific database or ERP system.