With the digital transformation of smart grids, power material management uses massive heterogeneous data, including structured procurement tables and unstructured demand texts. Existing multimodal models like TAMO fuse tabular and textual modalities to solve data silos, yet their static hard fusion brings extra noise. This paper proposes AdaTAMO, a prior-guided adaptive framework based on TAMO, with three core contributions. At the model level, it designs a domain knowledge-driven adaptive gating mechanism, using heuristic semantic rules to dynamically fuse text and table modalities on demand, reducing noise. At the data level, it builds Power-TableQA, a dedicated dataset for power material reasoning; the full dataset is private for compliance, but its construction pipeline and prompt templates are open-sourced for reproducibility. At the application level, it presents power grid material management scenarios and clarifies the model’s integration path. Experiments show that AdaTAMO performs comparably or better on general datasets, and outperforms baselines on the domain dataset, with higher query accuracy and interpretability for material demand decision making.
Yang et al. (Sat,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: