How to assess quality of BMF algorithms?
Radim Bělohlávek, Jan Outrata, Martin Trnečka · 2016
We critically examine the problem of quality assessment of algorithms for Boolean matrix factorization. We argue that little attention is paid to this problem in the literature. We view this problem as a multifaceted one and identify key aspects with respect to which the quality of algorithms should be assessed. Because of its utmost importance, we focus on assessment of quality of sets of factors extracted from Boolean data, propose ways to assess such quality and provide experimental evaluation involving selected factorization algorithms. We argue that the views involved in our proposal, represent reasonable basic standpoints for further systematic approaches to quality assessment.