Last decades have witnessed a radical change in the way information spreads. Social networks provide a constantly updated pool of news to the end-users but the absence of systematic control and moderation of the posts easily leads to spread unverified news with an instrumental value and likely to be dangerous, which are referred to as rumours. To tackle this issue various systems for automatic rumour detection among conversations, i.e. an aggregated set of posts, have been recently presented in the literature. However, few efforts have been directed towards rumour detection at the level of single posts (micro-level), which is the challenging scenario that we tackle in this work. Moving at a finer scale is an urgent need since both rumour and non-rumour posts are included in the same conversation. Here the rumour detection issue is addressed presenting a novel feature selection approach, which characterizes the feature space aiming at minimizing samples in unreliable configurations. This approach is compared with other state-of-the-art methods using a pool of different learning algorithms on two health-related Twitter datasets, labelled at the micro-level. Our proposal yields promising results: it outperforms other feature selection approaches with a best accuracy of 96.8% and enhances the performance of our previous work up to 5%. These findings prove the potential of the feature selection method introduced, which gives access to samples distribution in the feature space, providing privileged information for the construction of the classifier decision boundaries. Nonetheless they also bring a step forward the micro-level rumour detection analysis.

Rule-based Space Characterization for Rumour Detection in Health

Sicilia R
;
Merone M;Soda P
2021-01-01

Abstract

Last decades have witnessed a radical change in the way information spreads. Social networks provide a constantly updated pool of news to the end-users but the absence of systematic control and moderation of the posts easily leads to spread unverified news with an instrumental value and likely to be dangerous, which are referred to as rumours. To tackle this issue various systems for automatic rumour detection among conversations, i.e. an aggregated set of posts, have been recently presented in the literature. However, few efforts have been directed towards rumour detection at the level of single posts (micro-level), which is the challenging scenario that we tackle in this work. Moving at a finer scale is an urgent need since both rumour and non-rumour posts are included in the same conversation. Here the rumour detection issue is addressed presenting a novel feature selection approach, which characterizes the feature space aiming at minimizing samples in unreliable configurations. This approach is compared with other state-of-the-art methods using a pool of different learning algorithms on two health-related Twitter datasets, labelled at the micro-level. Our proposal yields promising results: it outperforms other feature selection approaches with a best accuracy of 96.8% and enhances the performance of our previous work up to 5%. These findings prove the potential of the feature selection method introduced, which gives access to samples distribution in the feature space, providing privileged information for the construction of the classifier decision boundaries. Nonetheless they also bring a step forward the micro-level rumour detection analysis.
File in questo prodotto:
File Dimensione Formato  
PublishedEAAI.pdf

solo utenti autorizzati

Descrizione: Last decades have witnessed a radical change in the way information spreads. Social networks provide a constantly updated pool of news to the end-users but the absence of systematic control and moderation of the posts easily leads to spread unverified news with an instrumental value and likely to be dangerous, which are referred to as rumours. To tackle this issue various systems for automatic rumour detection among conversations, i.e. an aggregated set of posts, have been recently presented in the literature. However, few efforts have been directed towards rumour detection at the level of single posts (micro-level), which is the challenging scenario that we tackle in this work. Moving at a finer scale is an urgent need since both rumour and non-rumour posts are included in the same conversation. Here the rumour detection issue is addressed presenting a novel feature selection approach, which characterizes the feature space aiming at minimizing samples in unreliable configurations. This approach is compared with other state-of-the-art methods using a pool of different learning algorithms on two health-related Twitter datasets, labelled at the micro-level. Our proposal yields promising results: it outperforms other feature selection approaches with a best accuracy of 96.8% and enhances the performance of our previous work up to 5%. These findings prove the potential of the feature selection method introduced, which gives access to samples distribution in the feature space, providing privileged information for the construction of the classifier decision boundaries. Nonetheless they also bring a step forward the micro-level rumour detection analysis.
Tipologia: Abstract
Licenza: Copyright dell'editore
Dimensione 2.6 MB
Formato Adobe PDF
2.6 MB Adobe PDF   Visualizza/Apri   Richiedi una copia

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/20.500.12610/5750
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 5
  • ???jsp.display-item.citation.isi??? ND
social impact