The experimental landscape in natural language processing for social media is too fragmented. Each year, new shared tasks and datasets are proposed, ranging from classics like sentiment analysis to irony detection or emoji prediction. Therefore, it is unclear what the current state of the art is, as there is no standardized evaluation protocol, neither a strong set of baselines trained on such domainspecific data. In this paper, we propose a new evaluation framework (TWEETEVAL) consisting of seven heterogeneous Twitter-specific classification tasks. We also provide a strong set of baselines as starting point, and compare different language modeling pre-training strategies. Our initial experiments show the effectiveness of starting off with existing pretrained generic language models, and continue training them on Twitter corpora.

Emojis are ideograms which are naturally combined with plain text to visually complement or condense the meaning of a message. Despite being widely used in social media, their underlying semantics have received little attention from a Natural Language Processing standpoint. In this paper, we investigate the relation between words and emojis, studying the novel task of predicting which emojis are evoked by text-based tweet messages. We train several models based on Long ShortTerm Memory networks (LSTMs) in this task. Our experimental results show that our neural model outperforms two baselines as well as humans solving the same task, suggesting that computational models are able to better capture the underlying semantics of emojis.

Automatic detection of figurative language is a challenging task in computational linguistics. Recognising both literal and figurative meaning is not trivial for a machine and in some cases it is hard even for humans. For this reason novel and accurate systems able to recognise figurative languages are necessary. We present in this paper a novel computational model capable to detect sarcasm in the social network Twitter (a popular microblogging service which allows users to post short messages). Our model is easy to implement and, unlike previous systems, it does not include patterns of words as features. Our seven sets of lexical features aim to detect sarcasm by its inner structure (for example unexpectedness, intensity of the terms or imbalance between registers), abstracting from the use of specific terms.

Pubblicazione resa disponibile nei termini della licenza Creative Commons Attribuzione -Non commerciale -Non opere derivate 4.0

No abstract