Culinary Crossroads: A RAG Framework for Enhancing Diversity in Cross-Cultural Recipe Adaptation

TL;DR

The paper confronts the challenge of producing diverse, culturally grounded recipe adaptations in cross-cultural contexts. It shows that standard Retrieval Augmented Generation (RAG) struggles to harness contextual diversity for varied outputs. To address this, it introduces CARRIAGE, a plug-and-play framework with query rewriting, diversity-aware re-ranking, dynamic context organization, and contrastive context injection that collectively enhance diversity and preserve quality. Experimental results on the RecetasDeLaAbuel@ dataset demonstrate Pareto-efficient improvements over baselines, highlighting the framework's potential to support multiple user preferences in culturally aware culinary generation.

Abstract

In cross-cultural recipe adaptation, the goal is not only to ensure cultural appropriateness and retain the original dish's essence, but also to provide diverse options for various dietary needs and preferences. Retrieval Augmented Generation (RAG) is a promising approach, combining the retrieval of real recipes from the target cuisine for cultural adaptability with large language models (LLMs) for relevance. However, it remains unclear whether RAG can generate diverse adaptation results. Our analysis shows that RAG tends to overly rely on a limited portion of the context across generations, failing to produce diverse outputs even when provided with varied contextual inputs. This reveals a key limitation of RAG in creative tasks with multiple valid answers: it fails to leverage contextual diversity for generating varied responses. To address this issue, we propose CARRIAGE, a plug-and-play RAG framework for cross-cultural recipe adaptation that enhances diversity in both retrieval and context organization. To our knowledge, this is the first RAG framework that explicitly aims to generate highly diverse outputs to accommodate multiple user preferences. Our experiments show that CARRIAGE achieves Pareto efficiency in terms of diversity and quality of recipe adaptation compared to closed-book LLMs.

Figures (6)

• Figure 1: A typical example of diverse choices in cross-cultural recipe adaptation. Nopal is a common ingredient in Mexican cuisine. However, adapting it to Spanish culinary preferences requires certain modifications, and there are many different alternatives (highlighted in green) to consider. This illustrates the necessity for our model to account for diversity in the adaptation process.
• Figure 2: Overview of CARRIAGE. Diversity components are highlighted. We first enhance the diversity of retrieved results, then we enable more diverse use of contextual information via dynamic context selection, and inject contrastive context to prevent the LLM from generating outputs similar to previously generated recipes.
• Figure 3: Trade-offs between diversity, cultural appropriateness, and source preservation for CARRIAGE and two top-performing LLMs under different temperature settings. Compared to Closed-book Qwen2.5 and LLaMA3.1 baselines, our model shows better Pareto efficiency across both key trade-offs.
• Figure 4: Pearson correlation matrix between metrics.
• Figure 5: Global ingredient metric across all the inputs. All adaptation methods reduce diversity compared to the original recipes, especially LLM and RAG, which notably increase the use of high-frequency ingredients.