AdhesiveTableQA: A Real-World Complex Table QA Dataset

We present AdhesiveTableQA, a specialized benchmark for Table Question Answering (TQA) focused on real-world adhesive material selection. The dataset includes 816 human-verified natural language questions with cell-level answers drawn from technical tables featuring complex and diverse layouts. Developed in collaboration with domain experts, it reflects industrial scenarios where precise table interpretation is essential. By targeting a practical application, this benchmark contributes to the ongoing effort to establish a standardized dataset annotation format for Unifying Table Understanding (TU) in real industrial applications. AdhesiveTableQA also serves as a valuable resource for assessing fine-grained knowledge extraction from (semi-)structured data, advancing TU in industrial settings.