IS@Bentley Seminar - Mickael Ringeval

Jingle, Jangle, and Construct Drift: A Human-AI Approach to Semantic Diagnosis Abstract: Constructs lose semantic coherence when their labels, definitions, or intended operationalizations no longer align or become difficult to distinguish from neighboring constructs. We develop the Label-Definition-Measure-Item framework and the Construct Diagnosis Engine (CDE) to identify where such problems originate. CDE uses GenAI to extract measures from full-text articles and semantic embeddings to compare construct representations within a shared space. Semantic Diagnostic Indicators assess alignment across construct layers, overall within-construct coherence, and between-construct distinctiveness. An application to five related IT-use constructs shows that closely related labels can conceal different conceptualizations and operationalizations. It also shows that similar diagnostic signals may reflect either expected theoretical structure or candidate semantic drift. CDE does not automate validity judgments. It provides a reproducible, multi-layer diagnostic profile that directs researcher interpretation and construct refinement.