Cross-Platform Evaluation for Italian Hate Speech Detection

English. Despite the number of approaches recently proposed in NLP for detecting abusive language on social networks , the issue of developing hate speech detection systems that are robust across different platforms is still an unsolved problem. In this paper we perform a comparative evaluation on datasets for hate speech detection in Italian, extracted from four different social media platforms, i.e. Facebook, Twitter, Instagram and What-sApp. We show that combining such platform-dependent datasets to take advantage of training data developed for other platforms is beneficial, although their impact varies depending on the social network under consideration. 1 Italiano. Nonostante si osservi un cre-scente interesse per approcci che identi-fichino il linguaggio offensivo sui social network attraverso l'NLP, la necessita di sviluppare sistemi che mantengano una buona performance anche su piattaforme diversee ancora un tema di ricerca aper-to. In questo contributo presentiamo una valutazione comparativa su dataset per l'identificazione di linguaggio d'odio pro-venienti da quattro diverse piattaforme: Facebook, Twitter, Instagram and Wha-tsApp. Lo studio dimostra che, combinan-do dataset diversi per aumentare i dati di training, migliora le performance di clas-sificazione, anche se l'impatto varia a se-conda della piattaforma considerata. 1

[1]  Paolo Rosso,et al.  Overview of the Task on Automatic Misogyny Identification at IberEval 2018 , 2018, IberEval@SEPLN.

[2]  Tomas Mikolov,et al.  Enriching Word Vectors with Subword Information , 2016, TACL.

[3]  Prakhar Gupta,et al.  Learning Word Vectors for 157 Languages , 2018, LREC.

[4]  Felice Dell'Orletta,et al.  Multi-task Learning in Deep Neural Networks at EVALITA 2018 , 2018, EVALITA@CLiC-it.

[5]  Viviana Patti,et al.  Hurtlex: A Multilingual Lexicon of Words to Hurt , 2018, CLiC-it.

[6]  Yoshua Bengio,et al.  Learning Phrase Representations using RNN Encoder–Decoder for Statistical Machine Translation , 2014, EMNLP.

[7]  Michael Wiegand,et al.  Overview of the GermEval 2018 Shared Task on the Identification of Offensive Language , 2018 .

[8]  Serena Villata,et al.  A System to Monitor Cyberbullying based on Message Classification and Social Network Analysis , 2019, Proceedings of the Third Workshop on Abusive Language Online.

[9]  Dirk Hovy,et al.  Proceedings of the First Workshop on Abusive Language Online , 2017 .

[10]  Nikos Pelekis,et al.  DataStories at SemEval-2017 Task 4: Deep LSTM with Attention for Message-level and Topic-based Sentiment Analysis , 2017, *SEMEVAL.

[11]  Sara Tonelli,et al.  Creating a WhatsApp Dataset to Study Pre-teen Cyberbullying , 2018, ALW.

[12]  Felice Dell'Orletta,et al.  Overview of the EVALITA 2018 Hate Speech Detection Task , 2018, EVALITA@CLiC-it.

[13]  Malvina Nissim,et al.  Sentiment analysis on Italian tweets , 2013, WASSA@NAACL-HLT.

[14]  Serena Villata,et al.  Comparing Different Supervised Approaches to Hate Speech Detection , 2018, EVALITA@CLiC-it.