Learning enhanced representations for tabular data via neighborhood propagation

Prediction over tabular data is an essential and fundamental problem in many important downstream tasks. However, existing methods either treat a data instance of the table independently as input or do not jointly utilize multi-row features and labels to directly change and enhance target data representations. In this paper, we propose to 1) construct a hypergraph from relevant data instance retrieval to model the cross-row and cross-column patterns of those instances, and 2) perform message Propagation to Enhance the target data instance representations for Tabular prediction tasks. Specifically, our tailored message propagation step benefits from both the fusion of label and features during propagation, as well as locality-aware high-order feature interactions. Experiments on two important tabular data prediction tasks validate the superiority of the proposed PET model relative to other baselines. Additionally, we demonstrate the effectiveness of the model components and the feature enhancement ability of PET via various ablation studies and visualizations. The code is available at https://github.com/KounianhuaDu/PET.