Juan Manuel Rodríguez

Training seeds and model-selection stability in recommender-system evaluation

It is suggested that reporting single-seed results can overstate the stability of recommender system evaluation, and that training seeds should be treated as part of the evaluation protocol.