The massive growth of user-generated web content through blogs, online forums and most notably, social media networks, led to a large spreading of hatred or abusive messages which have to be moderated. This paper proposes a supervised approach to hate speech detection towards immigrants and women in English tweets. Several models have been developed ranging from feature-engineering approaches to neural ones. We also carried out a detailed error analysis to show main causes of misclassification.

This paper describes the CASOAR corpus, the first manually annotated corpus that explores the impact of discourse structure on sentiment analysis with a study of movie reviews in French and in English as well as letters to the editor in French. While annotating opinions at the expression, the sentence or the document level is a well-established task and relatively straightforward, discourse annotation remains difficult, especially for non-experts. Therefore, combining both annotations poses several methodological problems that we address here. We propose a multi-layered annotation scheme that includes: the complete discourse structure according to the Segmented Discourse Representation Theory, the opinion orientation of elementary discourse units and opinion expressions, and their associated features. We detail each layer, explore the interactions between them and discuss our results. In particular, we examine the correlation between discourse and semantic category of opinion expressions, the impact of discourse relations on both subjectivity and polarity analysis and the impact of discourse on the determination of the overall opinion of a document. Our results demonstrate that discourse is an important cue for sentiment analysis, at least for the corpus genres we have studied.

We propose in this paper a supervised learning approach to identify discourse relations in Arabic texts. To our knowledge, this work represents the first attempt to focus on both explicit and implicit relations that link adjacent as well as non adjacent Elementary Discourse Units (EDUs) within the Segmented Discourse Representation Theory (SDRT). We use the Discourse Arabic Treebank corpus (D-ATB) which is composed of newspaper documents extracted from the syntactically annotated Arabic Treebank v3.2 part3 where each document is associated with complete discourse graph according to the cognitive principles of SDRT. Our list of discourse relations is composed of a three-level hierarchy of 24 relations grouped into 4 top-level classes. To automatically learn them, we use state of the art features whose efficiency has been empirically proved. We investigate how each feature contributes to the learning process. We report our experiments on identifying fine-grained discourse relations, mid-level classes and also top-level classes. We compare our approach with three baselines that are based on the most frequent relation, discourse connectives and the features used by Al-Saif and Markert (2011). Our results are very encouraging and outperform all the baselines with an F-score of 78.1% and an accuracy of 80.6%. ª 2014 The Authors. Production and hosting by Elsevier B.V. on behalf of King Saud University. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/3.0/).

International audienceIn this article, we propose the first work that investigates the feasibility of Arabic discourse segmentation into elementary discourse units within the segmented discourse representation theory framework. We first describe our annotation scheme that defines a set of principles to guide the segmentation process. Two corpora have been annotated according to this scheme: elementary school textbooks and newspaper documents extracted from the syntactically annotated Arabic Treebank. Then, we propose a multiclass supervised learning approach that predicts nested units. Our approach uses a combination of punctuation, morphological, lexical, and shallow syntactic features. We investigate how each feature contributes to the learning process. We show that an extensive morphological analysis is crucial to achieve good results in both corpora. In addition, we show that adding chunks does not boost the performance of our system

No abstract