RiCoRecA is a novel annotation schema for parsing cooking recipes into workflow representations suitable for IoT automation. The schema combines named entity recognition, relation classification, and coreference resolution into a single joint model. A dataset of 156 annotated recipes was created, and experiments comparing PEGASUS-X and LongT5 transformer models showed PEGASUS-X outperformed LongT5 by 39% in F-Score across all tasks, achieving near-human performance. The approach differs from previous work by handling conditional statements for sensor-based actions and learning multiple information extraction tasks jointly rather than through pipelines.
Table of contents
1 Introduction2 Related studies3 Annotation schema4 Dataset annotation5 Generative encoder-decoder transformer model6 Experiments7 Discussion and future work8 ConclusionData availability statementAuthor contributionsFundingConflict of interestGenerative AI statementPublisher's noteSupplementary materialFootnotesReferences341 Impressions