Abstract
Clustering plays a pivotal role in exploratory data analysis, especially
in scenarios where prior knowledge about a subset of samples is
available, prompting a rapid development of Semi-supervised clustering
approaches. In this work, we introduce a novel method for seamlessly
integrating prior knowledge into the widely recognized Normalized Cut
Clustering (NCC) algorithm, thereby offering a natural extension within
the realm of semi-supervised clustering. Similarly to NCC, the solution
of the proposed method is spectral-based in the form of an inhomogeneous
eigenvalue problem. Also, we show that the proposed method can be seen
as a generalization of NCC, enhancing its applicability in scenarios
where prior information is available. In addition, we design an adequate
and comprehensive experimental setup. This setup not only tests the
competitiveness of our approach but also showcases its superiority in
terms of performance metrics.