In this contribution we describe the system(i.e. a statistical model) used to participatein Evalita conference 2020, SardiStance(Tasks A and B) and Haspeede2 (TasksA and B). We first developed a classifierby extracting features from the texts andthe social network of users. Then, wefit the data through an extreme gradientboosting, with cross-validation tuning ofthe hyper-parameters. A key factor for agood performance in SardiStance Task Bwas the features extraction by using Mul-tidimensional Scaling of the distance ma-trix (minimum path, undirected graph) ap-plied on each network. The second sys-tem exploits the same features above, butit trains and performs predictions in two-steps.The performances proved to belower than those of the single-step model.
TextWiller @ SardiStance, HaSpeede2: Text or Con-text? A Smart Use of Social Network Data in Predicting Polarization / Ferraccioli, Federico; Sciandra, Andrea; Da Pont, Mattia; Girardi, Paolo; Solari and Livio Finos, Dario. - (2020). (Intervento presentato al convegno EVALITA 2020 - Seventh Evaluation Campaign of Natural Language Processing and Speech Tools for Italian tenutosi a Online event nel December 17th, 2020).
TextWiller @ SardiStance, HaSpeede2: Text or Con-text? A Smart Use of Social Network Data in Predicting Polarization
Andrea Sciandra;
2020
Abstract
In this contribution we describe the system(i.e. a statistical model) used to participatein Evalita conference 2020, SardiStance(Tasks A and B) and Haspeede2 (TasksA and B). We first developed a classifierby extracting features from the texts andthe social network of users. Then, wefit the data through an extreme gradientboosting, with cross-validation tuning ofthe hyper-parameters. A key factor for agood performance in SardiStance Task Bwas the features extraction by using Mul-tidimensional Scaling of the distance ma-trix (minimum path, undirected graph) ap-plied on each network. The second sys-tem exploits the same features above, butit trains and performs predictions in two-steps.The performances proved to belower than those of the single-step model.
Pubblicazioni consigliate
I metadati presenti in IRIS UNIMORE sono rilasciati con licenza Creative Commons CC0 1.0 Universal, mentre i file delle pubblicazioni sono rilasciati con licenza Attribuzione 4.0 Internazionale (CC BY 4.0), salvo diversa indicazione.
In caso di violazione di copyright, contattare Supporto Iris
