RiCoRecA: rich cooking recipe annotation schema.

Area of Science:

• Artificial Intelligence
• Human-Computer Interaction
• Computational Linguistics

Background:

• Modern smart kitchens lack integrated functionality, with devices operating in isolation.
• The vision of an ambient kitchen with automated, interconnected devices remains unrealized.

Purpose of the Study:

• To introduce RiCoRecA, a novel schema for parsing cooking recipes into machine-readable workflows.
• To enable the automation of kitchen tasks and advance the concept of ambient kitchens.

Main Methods:

• Developed a schema requiring named entity annotation, relation identification, coreference resolution, and entity tracking.
• Employed a joint model to learn multiple information extraction tasks simultaneously.
• Created a dataset annotated according to the RiCoRecA schema.
• Compared PEGASUS-X and LongT5 transformer models for recipe-to-workflow parsing.

Main Results:

• The PEGASUS-X model significantly outperformed LongT5 across all annotation tasks.
• PEGASUS-X achieved a 39% higher average F-Score compared to LongT5.
• The PEGASUS-X model demonstrated performance comparable to human annotators.

Conclusions:

• RiCoRecA provides a viable schema for representing recipes as workflows for automation.
• Joint learning of information extraction tasks is effective for recipe parsing.
• PEGASUS-X shows exceptional promise for advancing smart kitchen automation through accurate recipe understanding.