This work is concerned with classifying Web job advertise- ments against a standard classification system of occupations, by apply- ing and comparing different text classification techniques. As a first step, we evaluated the classification algorithms using a hit/not-hit approach, that is either the prediction is correct or not compared to a gold classi- fication provided by domain experts. Then, we built a distance function on top of the affinity relationship between occupations provided by the classification system. Both the classification scores we computed and the affinity distance employed have allowed a more finely grained evaluation of the classified outcomes, providing to authors useful insights towards the improvement of the classification process.
Classification of web job advertisements: A case study / Amato, Flora; Boselli, Roberto; Cesarini, Mirko; Mercorio, Fabio; Mezzanzanica, Mario; Moscato, Vincenzo; Persia, Fabio; Picariello, Antonio. - (2015), pp. 144-151. (Intervento presentato al convegno 23rd Italian Symposium on Advanced Database Systems, SEBD 2015 tenutosi a Gaeta (Italy) nel June 14-17, 2015).
Classification of web job advertisements: A case study
AMATO, FLORA;MOSCATO, VINCENZO;PICARIELLO, ANTONIO
2015
Abstract
This work is concerned with classifying Web job advertise- ments against a standard classification system of occupations, by apply- ing and comparing different text classification techniques. As a first step, we evaluated the classification algorithms using a hit/not-hit approach, that is either the prediction is correct or not compared to a gold classi- fication provided by domain experts. Then, we built a distance function on top of the affinity relationship between occupations provided by the classification system. Both the classification scores we computed and the affinity distance employed have allowed a more finely grained evaluation of the classified outcomes, providing to authors useful insights towards the improvement of the classification process.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.