Calibration of the mass-decorrelated ParticleNet tagger for boosted $\mathrm{b}\bar{\mathrm{b}}$ and $\mathrm{c}\bar{\mathrm{c}}$ jets using LHC Run 2 data

The calibration of the new generation jet tagging algorithms exploiting advanced machine learning techniques becomes a challenging task. This note presents a novel approach for the calibration of the mass-decorrelated ParticleNet (ParticleNet-MD) boosted jet flavour tagging algorithm, focusing on the $\mathrm{X} \rightarrow \mathrm{b}\bar{\mathrm{b}}$ and $\mathrm{X} \rightarrow \mathrm{c}\bar{\mathrm{c}}$ mode. The approach builds upon the already established method used for the calibration of the previous generation boosted jet taggers, i.e., using proxy jets from gluon splitting to a pair of bottom or charm quarks. However, new techniques have been introduced to improve the similarity between proxy and signal jets and control the systematic uncertainties associated with the corrections. Data-to-simulation scale factors are derived for the three data taking years of Run 2 with the CMS experiment, based on different working points.