Deep Learning Based Named Entity Recognition Models for Recipes

TL;DR

This paper develops and evaluates deep learning models for recognizing named entities in recipes, demonstrating that fine-tuned spaCy-transformer models outperform large language models with few-shot prompting across multiple datasets.

Contribution

It introduces a large, diverse dataset for recipe NER and systematically compares deep learning approaches, highlighting the effectiveness of fine-tuned models over LLM prompting.

Findings

Fine-tuned spaCy-transformer achieved macro-F1 scores above 95%.

Few-shot prompting on LLMs performed poorly for recipe NER.

Created a large, annotated dataset from recipe texts for improved NER training.

Abstract

Food touches our lives through various endeavors, including flavor, nourishment, health, and sustainability. Recipes are cultural capsules transmitted across generations via unstructured text. Automated protocols for recognizing named entities, the building blocks of recipe text, are of immense value for various applications ranging from information extraction to novel recipe generation. Named entity recognition is a technique for extracting information from unstructured or semi-structured data with known labels. Starting with manually-annotated data of 6,611 ingredient phrases, we created an augmented dataset of 26,445 phrases cumulatively. Simultaneously, we systematically cleaned and analyzed ingredient phrases from RecipeDB, the gold-standard recipe data repository, and annotated them using the Stanford NER. Based on the analysis, we sampled a subset of 88,526 phrases using a clustering-based approach while preserving the diversity to create the machine-annotated dataset. A thorough investigation of NER approaches on these three datasets involving statistical, fine-tuning of deep learning-based language models and few-shot prompting on large language models (LLMs) provides deep insights. We conclude that few-shot prompting on LLMs has abysmal performance, whereas the fine-tuned spaCy-transformer emerges as the best model with macro-F1 scores of 95.9%, 96.04%, and 95.71% for the manually-annotated, augmented, and machine-annotated datasets, respectively.