Quantum recommendation preprint reports comparable results with fewer parameters

Researchers report in a preprint that a quantum media recommendation model matched a classical BiLSTM model on a similarity-learning task using 47k parameters versus 17.8M. The evaluation used programme descriptions from one week of BBC iPlayer data. The system converts the semantic structure of programme synopses into trainable quantum circuits. When the classical model was restricted to the same parameter budget, its Mean Average Precision fell by 3.63 percentage points. Researchers also ran circuits on Quantinuum's System Model H2 and its emulator, reporting that the similarity signal survived noisy conditions. The findings establish parameter efficiency in this setting, with evaluation limited to that dataset and task.