Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

FoodTaxo: Generating Food Taxonomies with Large Language Models

Pascal Wullschleger, Majid Zarharan, Donnacha Daly, Marc Pouly, Jennifer Foster

TL;DR

FoodTaxo investigates automated taxonomy generation and completion for the food-technology domain using Large Language Models. The approach employs seed-based completion and seed-free generation in an iterative, bottom-up framework with prompting, retrieval, and NLI-based verification, augmented by backtracking to enforce structural validity. Empirical results across five taxonomies show the LLM-based methods can match or exceed state-of-the-art completion on several datasets and produce competitive, reference-free assessments for generated taxonomies, though placement of non-leaf concepts remains challenging. The work demonstrates potential for scalable, plug-and-play taxonomy construction beyond food, but highlights the need for improved non-leaf placement accuracy and further methodological refinements to reach production-level utility.

Abstract

We investigate the utility of Large Language Models for automated taxonomy generation and completion specifically applied to taxonomies from the food technology industry. We explore the extent to which taxonomies can be completed from a seed taxonomy or generated without a seed from a set of known concepts, in an iterative fashion using recent prompting techniques. Experiments on five taxonomies using an open-source LLM (Llama-3), while promising, point to the difficulty of correctly placing inner nodes.

FoodTaxo: Generating Food Taxonomies with Large Language Models

TL;DR

FoodTaxo investigates automated taxonomy generation and completion for the food-technology domain using Large Language Models. The approach employs seed-based completion and seed-free generation in an iterative, bottom-up framework with prompting, retrieval, and NLI-based verification, augmented by backtracking to enforce structural validity. Empirical results across five taxonomies show the LLM-based methods can match or exceed state-of-the-art completion on several datasets and produce competitive, reference-free assessments for generated taxonomies, though placement of non-leaf concepts remains challenging. The work demonstrates potential for scalable, plug-and-play taxonomy construction beyond food, but highlights the need for improved non-leaf placement accuracy and further methodological refinements to reach production-level utility.

Abstract

We investigate the utility of Large Language Models for automated taxonomy generation and completion specifically applied to taxonomies from the food technology industry. We explore the extent to which taxonomies can be completed from a seed taxonomy or generated without a seed from a set of known concepts, in an iterative fashion using recent prompting techniques. Experiments on five taxonomies using an open-source LLM (Llama-3), while promising, point to the difficulty of correctly placing inner nodes.

Paper Structure

This paper contains 47 sections, 4 equations, 2 figures, 9 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. Problem Definition
  5. Completing Taxonomies
  6. Generating Taxonomies
  7. Prompts
  8. Verifying Placements
  9. Backtracking
  10. Completion Experiments
  11. Data
  12. Evaluation
  13. Model selection
  14. Hypothesis testing
  15. Results
  16. ...and 32 more sections

Figures (2)

  • Figure 1: The basic intuition behind the generation process. We start with a set of known concepts and iteratively construct a taxonomy in a bottom-up procedure by prompting .
  • Figure 2: Examples of generated sub-graphs of the taxonomies. Depiction (a) is the gold standard neighborhood of coffee, while (b) is an example of the generated taxonomy based on SemEval-Food leaves. An erroneously classified non-leaf is shown in (c).