SHARE: a System for Hierarchical Assistive Recipe Editing

Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 11077 - 11090 December 7-11, 2022 ©2022 Association for Computational Linguistics SHARE: a System for Hierarchical Assistive Recipe Editing Shuyang Li UC San Diego lishuyang@meta.com Yufei Li UC Riverside yli927@ucr.edu Jianmo Ni UC San Diego jianmon@google.com Julian McAuley UC San Diego jmcauley@ucsd.edu Abstract The large population of home cooks with di- etary restrictions is under-served by existing cooking resources and recipe generation mod- els. To help them, we propose the task of con- trollable recipe editing: adapt a base recipe to satisfy a user-specified dietary constraint. This task is challenging, and cannot be ad- equately solved with human-written ingredi- ent substitution rules or existing end-to-end recipe generation models. We tackle this prob- lem with SHARE: a System for Hierarchical Assistive Recipe Editing, which performs si- multaneous ingredient substitution before gen- erating natural-language steps using the edited ingredients. By decoupling ingredient and step editing, our step generator can explicitly inte- grate the available ingredients. Experiments on the novel RecipePairs dataset—83K pairs of similar recipes where each recipe satisfies one of seven dietary constraints—demonstrate that SHARE produces convincing, coherent recipes that are appropriate for a target dietary con- straint. We further show through human evalua- tions and real-world cooking trials that recipes edited by SHARE can be easily followed by home cooks to create appealing dishes. 1 Introduction Cooking has played an integral role in human civi- lization and evolution for over 1.8 million years (Wrangham, 2009). A growing population fol- lows some form of dietary restriction (Goldberg et al., 2017), with motivations ranging from so- cioeconomic to medical (Shepherd et al., 1996; De- Salvo et al., 2016). Home cooks browsing recipes on online recipe aggregators may often encounter an interesting recipe that does not fit their dietary needs (e.g. vegetarianism), and would benefit from a way to edit that recipe to fit their needs. These restrictions are easy for users to specify, but exist- ing recipe websites offer few options for users with dietary constraints—even those with common re- strictions like gluten intolerance (Table 1). We see an opportunity to build a system to tailor recipes to fit users’ preferences and dietary constraints. To help such users, we introduce the task of con- trollable recipe editing: editing a base recipe so that it satisfies a user-specified dietary constraint (Figure 1). Controllable recipe editing can help home cooks of any experience level find diverse ways to satisfy their dietary needs and help develop interactive, personalized products like meal kits to accommodate individual needs and preferences. This is a challenging task: rule-based substitution methods and existing recipe generators cannot ade- quately account for both the dietary and structural impacts of ingredient substitutions. To tackle this task, we propose a System for Hierarchical Assistive Recipe Editing (SHARE). We first use a Transformer (Vaswani et al., 2017) encoder-decoder to perform multiple simultaneous ingredient substitutions conditioned on a dietary re- striction. We next employ a Transformer language model to write new instructions conditioned …

Content truncated to 500 words