Abstract
In the last years, a wide range of new similarity measures has been designed and applied to different contexts. Currently, there’s a deep lack in the validation and evaluation steps for novel similarities. In general, new measures are validated mostly through numerical experiments using different data sets. But this is not enough, as in order to gain relevance for a domain, a more complex validation process is necessary. To overcome this, a validation framework is proposed for the recently introduced ARP (Attractiveness-Relevance-Popularity) similarity measure. The validation process consists of five main steps: metrics conditions check, usefulness, expressivity, correlations to other measures, and noise robustness.
Citare
@Inproceedings{Limboi2021AVF,
author = {Sergiu Limboi and Mara Deac-Petrusel},
booktitle = {International Conference on Machine Learning and Applications},
title = {A Validation Framework for ARP Similarity Measure},
year = {2021}
}