Epicure: Multidimensional Flavor Structure in Food Ingredient Embeddings

TL;DR

This paper demonstrates that flavor, texture, and cultural knowledge are embedded in high-dimensional ingredient representations, which can be systematically extracted and classified into multiple meaningful dimensions.

Contribution

It introduces a method to recover and classify multidimensional flavor information from ingredient embeddings using an LLM-augmented curation pipeline.

Findings

Identified at least fifteen classifiable flavor and cultural dimensions.

Strengthened ingredient representations through canonicalization.

Revealed systematic encoding of culinary knowledge in embeddings.

Abstract

A chef's intuition about flavor, texture, and cultural identity represents tacit knowledge that is difficult to articulate yet central to culinary practice. We show that this knowledge is already encoded in FlavorGraph's 300-dimensional ingredient embeddings, trained on recipe cooccurrence and food chemistry, and that it can be systematically recovered. An LLM-augmented curation pipeline consolidates 6,653 raw FlavorGraph ingredients into 1,032 canonical entries, substantially strengthening the recoverable structure. We identify at least fifteen independently classifiable dimensions spanning taste, texture, geography, food processing, and culture.