ReciFineGold Dataset

ReciFineGold is a finely annotated recipe dataset designed to support research in controllable recipe generation, structured recipe understanding, and information extraction from procedural cooking text. The dataset contains step-by-step cooking instructions drawn from 500 real-world recipes, each manually annotated with span-level semantic labels that capture the key components of culinary actions, ingredients, tools, and states.

Each recipe in the dataset consists of:

• A unique recipe identifier (id)
• A natural language cooking procedure (text)
• A list of character-level annotations (label), where each annotation specifies:
  • the start and end offsets of a text span, and
  • its corresponding semantic entity type

Performance of Encoder Models Trained on The ReciFineGold Dataset

Entity Types

The dataset defines the following semantic entity types that were adopted from the Recipe Case Study Paper, each representing a different aspects of the cooking process:

Entity Type	Definition
FOOD	Edible items, including both raw ingredients and intermediate products
TOOL	Cooking tools such as knives, bowls, pans
DURATION	Time durations in cooking (e.g., 20 minutes)
QUANTITY	Quantities associated with ingredients
ACTION_BY_CHEF	Verbs for deliberate cook actions (e.g., bring in “Bring the mixture to a boil”)
ACTION_BY_CHEF_DISCONTINUOUS	Non-contiguous parts of compound chef actions (e.g., to a boil)
ACTION_BY_FOOD	Verbs where food is the agent (e.g., melt, boil)
ACTION_BY_TOOL	Verbs denoting tool actions (e.g., grind, beat)
FOOD_STATE	Descriptions of food’s state (e.g., chopped, soft)
TOOL_STATE	Descriptions of tool readiness (e.g., preheated, greased, covered)

How to use models trained on this dataset

Option 1 - Using the Knowledge Augmented and Entity Type-Specific (KAES) Models

The ReciFine library provides a lightweight inference wrapper (ReciFineNER) that handles the KAES prompting and decoding.

# Install ReciFine (see repo for latest install instructions)
pip install https://github.com/nuhu-ibrahim/ReciFine/archive/refs/tags/V1.zip

from recifine.inferencing.inference import ReciFineNER

ner = ReciFineNER.from_pretrained(
    model="recipebert", # recipebert, reciperoberta
    task_formulation="knowledge_guided",
    knowledge_type="question", # question, entity_type, example, definition, all
)

text = "Add 2 cups of chopped onions and fry until golden."
prediction = ner.process_text(
    text,
    entity_type="FOOD" # ACTION_BY_CHEF, ACTION_BY_CHEF_DISCONTINUOUS, ACTION_BY_FOOD, ACTION_BY_TOOL, DURATION, FOOD, FOOD_STATE, QUANTITY, TOOL, TOOL_STATE
)

print(prediction)

Option 2 - Using the Traditional BIO-n NER

The ReciFine library provides a lightweight inference wrapper (ReciFineNER) that handles extraction and decoding.

pip install https://github.com/nuhu-ibrahim/ReciFine/archive/refs/tags/V1.zip

from recifine.inferencing.inference import ReciFineNER

ner = ReciFineNER.from_pretrained(
    model="recipebert", # recipebert, reciperoberta
    task_formulation="traditional",
)

text = "Add 2 cups of chopped onions and fry until golden ."

prediction = ner.process_text(
    text
)

print(prediction)

Quick links (documentation + notebook)

• ReciFine library (recommended): training scripts, inference wrappers, and full documentation
  https://github.com/nuhu-ibrahim/ReciFine/tree/main

• Colab (end-to-end usage demo):
  https://colab.research.google.com/drive/1CatH2YOhnOWf-VglprEgxONWRkrXRBXo?usp=sharing&authuser=1

Research Papers

Knowledge-Augmented and Entity Type-Specific Token Classification

The knowledge-augmented and entity type-specific token classification model architecture is described in the paper Knowledge Augmentation Enhances Token Classification for Recipe Understanding.

@inproceedings{
  title     = {Knowledge Augmentation Enhances Token Classification for Recipe Understanding},
  author    = {Ibrahim, Nuhu and Stevens, Robert and Batista-Navarro, Riza},
  booktitle = {EACL},
  year      = {2026}
}

ReciFine Datasets and Controllable Recipe Generation

The ReciFine, ReciFineGold and ReciFineGen datasets are described in the paper ReciFine: Finely Annotated Recipe Dataset for Controllable Recipe Generation.

@inproceedings{
  title   = {ReciFine: Finely Annotated Recipe Dataset for Controllable Recipe Generation},
  author  = {Ibrahim, Nuhu and Ravikumar, Rishi and Stevens, Robert and Batista-Navarro, Riza},
  booktitle = {EACL},
  year    = {2026}
}