Losses that Cook: Topological Optimal Transport for Structured Recipe Generation
This addresses the challenge of generating accurate and coherent cooking recipes, which is an incremental improvement for natural language generation in a specific domain.
The paper tackles the problem of generating structured cooking recipes by introducing a topological loss that represents ingredient lists as point clouds in embedding space, which significantly improves ingredient- and action-level metrics while other losses enhance time/temperature precision, with human preference analysis showing the model is preferred in 62% of cases.
Cooking recipes are complex procedures that require not only a fluent and factual text, but also accurate timing, temperature, and procedural coherence, as well as the correct composition of ingredients. Standard training procedures are primarily based on cross-entropy and focus solely on fluency. Building on RECIPE-NLG, we investigate the use of several composite objectives and present a new topological loss that represents ingredient lists as point clouds in embedding space, minimizing the divergence between predicted and gold ingredients. Using both standard NLG metrics and recipe-specific metrics, we find that our loss significantly improves ingredient- and action-level metrics. Meanwhile, the Dice loss excels in time/temperature precision, and the mixed loss yields competitive trade-offs with synergistic gains in quantity and time. A human preference analysis supports our finding, showing our model is preferred in 62% of the cases.