RecipeNet: A Hierarchical Transformer for Recipe Data

Forum topic · 小凯 · 2026-08-18

Summary

RecipeNet is a hierarchical Transformer architecture introduced to handle recipe-structured data found in materials synthesis, pharmaceutical formulation, and industrial manufacturing. In these domains, procedures are represented as ordered sequences of steps, each containing heterogeneous structured fields. Existing tabular learning methods typically flatten this structure into fixed-schema representations, which limits their ability to capture hierarchical field interactions and procedural dependencies across steps. RecipeNet addresses this by stacking Transformer encoders: one level models field-level interactions within each step, while another captures sequential dependencies across steps. Experiments on multiple recipe datasets and tasks show that RecipeNet consistently outperforms existing tabular models, demonstrating the value of jointly hierarchical and sequential modeling for recipe representation learning. The paper, by Pin-Yen Huang, Sachin Chhabra, and Prasanth Sai Gouripeddi, is available on arXiv as 2508.08537.

Paper Overview

Research Area: Machine Learning Authors: Pin-Yen Huang, Sachin Chhabra, Prasanth Sai Gouripeddi Published: 2026-08-17 arXiv: 2508.08537

Original Abstract

Recipe data arises in domains such as materials synthesis, pharmaceutical formulation, and industrial manufacturing, where procedures are represented as ordered sequences of steps containing heterogeneous structured fields. Existing tabular learning methods typically flatten this structure into fixed-schema representations, limiting their ability to capture hierarchical field interactions and procedural dependencies. We propose RecipeNet, a hierarchical Transformer architecture that encodes field-level interactions within each step and sequential dependencies across steps through stacked Transformer encoders. Experiments on multiple recipe datasets and tasks demonstrate that RecipeNet consistently outperforms existing tabular models, highlighting the value of hierarchical and sequential modeling for recipe representation learning.

Key Ideas

Problem: Recipe-like structured data (ordered steps with heterogeneous fields) is poorly handled by standard tabular methods that flatten schemas.
Approach: A hierarchical Transformer that models field-level interactions within each step and sequence-level dependencies across steps using stacked Transformer encoders.
Results: Consistent performance gains over existing tabular models across multiple recipe datasets and tasks.

---

*Auto-collected on 2026-08-18*