Home  | Publications | HKK+19

SIDEKICK: Linear Correlation Clustering With Supervised Background Knowledge

MCML Authors

Peer Kröger

Prof. Dr.

Principal Investigator

* Former Principal Investigator

Link to Profile Thomas Seidl PI Matchmaking

Thomas Seidl

Prof. Dr.

Director

Database Systems, Data Mining and AI

Abstract

While explainable AI (XAI) is gaining in popularity, other more traditional machine learning algorithms can also benefit from increased explainability. A semi-supervised approach to correlation clustering opens up a promising design space that might provide such explainability to correlation clustering algorithms. In this work, semi-supervised linear correlation clustering is defined as the task of finding arbitrary oriented subspace clusters using only a small sample of supervised background knowledge provided by a domain experts. This work describes a first foray into this novel approach and provides an implementation of a basic algorithm to perform this task. We have found that even a small amount of supervised background knowledge can significantly improve the quality of correlation clustering in general. With confidence it can be stated, the results of this work have the potential to inspire several more semi-supervised approaches to correlation clustering in the future.

inproceedings

SISAP 2019

12th International Conference on Similarity Search and Applications. Newark, New York, USA, Oct 02-04, 2019.

Authors

M. A. X. Hünemörder • D. Kazempour • P. Kröger • T. Seidl

Links

DOI

Research Area

 A3 | Computational Models

BibTeXKey: HKK+19

Back to Top