Explainable Process Recommendation by Contextual Grounded Dynamic Multimodal Process KG

This lecture explores the challenges and solutions for process recommendation systems that require compositional reasoning. Dive into how Dynamic Multimodal Process Knowledge Graphs (DMPKGs) provide a neurosymbolic framework for analyzing complex processes like food recipes and AI pipelines. Learn why traditional recommendation systems struggle with unstructured data and how DMPKGs overcome these limitations by combining neural networks for pattern recognition with structured knowledge representations. The presentation demonstrates how this approach enables modular entity inference, captures entity interactions, allows continuous knowledge updates, and stores multimodal data—all contributing to more accurate and explainable recommendations. Examine two practical applications: analyzing recipe suitability for dietary needs and recommending optimal AI pipelines for specific tasks and datasets.