Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

CookingSense: A Culinary Knowledgebase with Multidisciplinary Assertions

Donghee Choi, Mogan Gim, Donghyeon Park, Mujeen Sung, Hyunjae Kim, Jaewoo Kang, Jihun Choi

TL;DR

CookingSense presents a large-scale culinary knowledge base constructed from web data, scientific literature, and recipes, combined with dictionary- and LM-driven filtering to yield multidimensional culinary assertions. The authors also introduce FoodBench, a benchmark for evaluating culinary decision-support tasks and to assess knowledge augmentation for retrieval-augmented generation. The pipeline generates 34 million categorized assertions from 68 million inputs, and semantic categorization enables targeted retrieval for downstream tasks. Empirical results show that CookingSense consistently improves RAG performance across culinary QA, flavor, and cultural tasks, illustrating its potential to power practical culinary decision-support tools and future domain-specific chatbots.

Abstract

This paper introduces CookingSense, a descriptive collection of knowledge assertions in the culinary domain extracted from various sources, including web data, scientific papers, and recipes, from which knowledge covering a broad range of aspects is acquired. CookingSense is constructed through a series of dictionary-based filtering and language model-based semantic filtering techniques, which results in a rich knowledgebase of multidisciplinary food-related assertions. Additionally, we present FoodBench, a novel benchmark to evaluate culinary decision support systems. From evaluations with FoodBench, we empirically prove that CookingSense improves the performance of retrieval augmented language models. We also validate the quality and variety of assertions in CookingSense through qualitative analysis.

CookingSense: A Culinary Knowledgebase with Multidisciplinary Assertions

TL;DR

CookingSense presents a large-scale culinary knowledge base constructed from web data, scientific literature, and recipes, combined with dictionary- and LM-driven filtering to yield multidimensional culinary assertions. The authors also introduce FoodBench, a benchmark for evaluating culinary decision-support tasks and to assess knowledge augmentation for retrieval-augmented generation. The pipeline generates 34 million categorized assertions from 68 million inputs, and semantic categorization enables targeted retrieval for downstream tasks. Empirical results show that CookingSense consistently improves RAG performance across culinary QA, flavor, and cultural tasks, illustrating its potential to power practical culinary decision-support tools and future domain-specific chatbots.

Abstract

This paper introduces CookingSense, a descriptive collection of knowledge assertions in the culinary domain extracted from various sources, including web data, scientific papers, and recipes, from which knowledge covering a broad range of aspects is acquired. CookingSense is constructed through a series of dictionary-based filtering and language model-based semantic filtering techniques, which results in a rich knowledgebase of multidisciplinary food-related assertions. Additionally, we present FoodBench, a novel benchmark to evaluate culinary decision support systems. From evaluations with FoodBench, we empirically prove that CookingSense improves the performance of retrieval augmented language models. We also validate the quality and variety of assertions in CookingSense through qualitative analysis.
Paper Structure (27 sections, 3 figures, 8 tables)

This paper contains 27 sections, 3 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. KB Construction
  4. Requirements
  5. Data Sources
  6. Web Data
  7. Scientific Papers
  8. Recipes
  9. Creation of Assertions
  10. Removal of Non-Generic Assertions
  11. Removal of Irrelevant Assertions
  12. Ingredient and Food Names
  13. Terms Collected from AI Assistant
  14. Semantic Categorization
  15. Evaluation
  16. ...and 12 more sections

Figures (3)

  • Figure 1: Our main knowledgebase, CookingSense, is built upon diverse culinary knowledge sources. We assess the usefulness of CookingSense using the benchmark framework, FoodBench, in assessing the capability of decision-making in the culinary domain.
  • Figure 2: CookingSense examples. We present a selection of examples representing two culinary concepts, such as rice and kimchi, across various types in Table \ref{['table:CookingSense_categorization_results']}.
  • Figure 3: Pipeline of CookingSense KB construction.