The CMS experiment makes use of a large variety of algorithms to identify the origin of particle jets measured in the detector. Through the study of jet substructure properties, jets originating from quarks, gluons, W/Z/Higgs bosons, top quarks and pileup interactions are identified and categorized. We present new techniques based on machine learning approaches developed for the analysis of the data collected during the LHC Run 2 that significantly surpass the performances of classical taggers.