QMOF-Rec: A multi-objective recommender-system framework for quantum-chemical MOF candidate screening

Screening metal–organic frameworks from incomplete metadata requires explicit separation of available evidence, missing descriptors, and application preferences. QMOF-Rec provides a query-conditioned recommendation framework that separates masked utility, material-space diversity, candidate generation, and auxiliary explanation. Evaluation uses 20,372 QMOF records, five predefined application scenarios, ten seeds, shared pools of 100, and top-five output. LEA and WeightedSum achieve Rel@5 = 0.9004 and NDCG@5 = 1.0000; hybrid diversity is 0.0188 for WeightedSum, 0.0204 for LEA, and 0.0215 for MMR. Differential evolution yields diversity 0.0213. MMR is faster than LEA under matched top-five work. Relevance metrics use the optimized utility and therefore measure utility preservation rather than independent relevance. Descriptor checks characterize internal screening consistency; absent adsorption, porosity, experimental, and expert-usefulness evidence prevents claims of validated application performance.