Epicure: Navigating the Emergent Geometry of Food Ingredient Embeddings

2026-05-21 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, quick

Summary

Epicure is a new family of three sibling skip-gram ingredient embeddings, retrained from scratch on a vast multilingual recipe corpus comprising 4.14M recipes from 11 sources across seven languages, including English, Chinese, and German. An LLM-augmented pipeline normalized raw ingredient strings into 1,790 canonical entries. The models leverage two distinct graphs: a 203,508-edge ingredient-ingredient NPMI co-occurrence graph and an 80,019-edge typed FlavorDB ingredient-compound graph featuring 2,247 typed compound nodes across 15 categories. Three Metapath2Vec variants—Cooc, Chem, and Core—were developed with shared architecture but different random-walk schemas. Cooc uses only the co-occurrence graph, Chem uses only typed compound metapaths, and Core blends both, positioning each model uniquely on the chemistry-versus-recipe-context spectrum.

Key takeaway

For AI Scientists developing food-related recommendation systems or culinary AI, this work demonstrates a robust method for creating nuanced ingredient embeddings. You should consider integrating both recipe co-occurrence and chemical compound data to enhance model accuracy and interpretability. This approach allows you to move beyond simple co-occurrence, enabling more sophisticated predictions about flavor compatibility and ingredient substitutions. Explore the Core model variant for a balanced perspective.

Key insights

Food ingredient embeddings can integrate both recipe co-occurrence and chemical compound data for richer representations.

Principles

• Blending co-occurrence and chemistry enriches embeddings.
• LLMs can normalize complex, raw ingredient data.
• Metapath2Vec adapts to diverse graph structures.

Method

An LLM-augmented pipeline normalizes ingredients, then Metapath2Vec variants are applied to a blended graph of ingredient co-occurrence and typed FlavorDB ingredient-compound relationships.

In practice

• Develop ingredient recommendation systems.
• Analyze flavor profiles for new dishes.
• Identify ingredient substitutions based on chemistry.

Topics

• Ingredient Embeddings
• Metapath2Vec
• Recipe Analysis
• Natural Language Processing
• Food Chemistry
• Multilingual Data

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, NLP Engineer

Related on AIssential

• On GATE, Text and Social Media Analysis, and Detecting Misinformation Online — A multi-layer analytical pipeline effectively extracts and categorizes digital identity aspects of agrifood products from online discourse.
• Building Context-Aware Search in Python with LLM Embeddings + Metadata — Context-aware semantic search combines embedding-based similarity with metadata filtering for relevant results respecting contextual constraints.
• Geometry of Semantic Space: Comparative Study of Discrete and Continuous Models — Graph-based semantic models offer more interpretable and structured meaning representations than high-dimensional transformer embeddings.
• Rethinking Negative Sampling for Knowledge Distillation in Retrieval, A Distillation Recipe for Small Language Model Search Agents, and More! — Effective information retrieval advancements often stem from rethinking fundamental assumptions and optimizing data representation or model behavior.
• Exploring NLP Applications in Food Research: my ATRIUM TNA visit to the GATE group — NLP tools significantly enhance social media data analysis for consumer behavior and food research.
• IdioLink: Retrieving Meaning Beyond Words Across Idiomatic and Literal Expressions — Idiom understanding in retrieval requires semantic abstraction beyond lexical overlap.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.