Hate speech and abusive language have become a common phenomenon on Arabic social media. Automatic hate speech and abusive detection systems can facilitate the prohibition of toxic textual contents. The complexity, informality and ambiguity of the Arabic dialects hindered the provision of the needed resources for Arabic abusive/hate speech detection research. In this paper, we introduce the first publicly-available Levantine Hate Speech and Abusive (L-HSAB) Twitter dataset with the objective to be a benchmark dataset for automatic detection of online Levantine toxic contents. We, further, provide a detailed review of the data collection steps and how we design the annotation guidelines such that a reliable dataset annotation is guaranteed. This has been later emphasized through the comprehensive evaluation of the annotations as the annotation agreement metrics of Cohen's Kappa (k) and Krippendorff's alpha (α) indicated the consistency of the annotations.

In this paper, we describe our contribution in SemEval-2018 contest. We tackled task 1 "Affect in Tweets", subtask E-c "Detecting Emotions (multi-label classification)". A multilabel classification system Tw-StAR was developed to recognize the emotions embedded in Arabic, English and Spanish tweets. To handle the multi-label classification problem via traditional classifiers, we employed the binary relevance transformation strategy while a TF-IDF scheme was used to generate the tweets' features. We investigated using single and combinations of several preprocessing tasks to further improve the performance. The results showed that specific combinations of preprocessing tasks could significantly improve the evaluation measures. This has been later emphasized by the official results as our system ranked 3 rd for both Arabic and Spanish datasets and 14 th for the English dataset.

Social media platforms have been witnessing a significant increase in posts written in the Tunisian dialect since the uprising in Tunisia at the end of 2010. Most of the posted tweets or comments reflect the impressions of the Tunisian public towards social, economical and political major events. These opinions have been tracked, analyzed and evaluated through sentiment analysis systems. In the current study, we investigate the impact of several preprocessing techniques on sentiment analysis using two sentiment classification models: Supervised and lexicon-based. These models were trained on three Tunisian datasets of different sizes and multiple domains. Our results emphasize the positive impact of preprocessing phase on the evaluation measures of both sentiment classifiers as the baseline was significantly outperformed when stemming, emoji recognition and negation detection tasks were applied. Moreover, integrating named entities with these tasks enhanced the lexicon-based classification performance in all datasets and that of the supervised model in medium and small sized datasets.

No abstract