Semantalytics: Multi-modal Experimental Data Integration and Analysis through Semantic Alignment

Abstract

We describe an event-based semantics-driven framework for heterogeneous dataset integration, alignment and data analysis. The testbed for this semantalytics prototype was the pilot study of a multi-step experiment examining the impact of changes in a conference room air system on air quality, human cognition and human physiology. Experimental measurements from environment sensors, human physiological sensors, cognitive tests and saliva samples were collected at varying sample rates and time frames. The observed data represented a variety of data types, ranging from real-time (streaming) air-quality measures to questionnaires and biological test results. We applied linked data principles to the data using domain-specific and authoritative ontologies to create an integrated knowledge graph embodying the entire experiment. Utilizing available SPARQL extensions to the R analytics platform, we constructed a multidimensional tensor structured around per-subject observational events, thus semantically aligning the experimental data for analysis and modeling. Ultimately, the semantic integration and alignment of heterogeneous data helped provide insights across the different experimental data types by enabling a unified analysis of the different data modes unified through the “lens” of the event structure of the experiment. This work was conducted in partnership with a multidisciplinary team of research scientists, architects, mathematicians, engineers, undergraduate students, and graduate students.