Moral Stance Recognition and Polarity Classification from Twitter and Elicited Text

Proceedings of Recent Advances in Natural Language Processing, pages 1069–1075, Varna, Bulgaria, Sep 2–4, 2019. https://doi.org/10.26615/978-954-452-056-4_123 1069 Moral Stance Recognition and Polarity Classification from Twitter and Elicited Text Wesley Ramos dos Santos University of S˜ao Paulo S˜ao Paulo, Brazil wesley.ramos.santos@usp.br Ivandr´e Paraboni University of S˜ao Paulo S˜ao Paulo, Brazil ivandre@usp.br Abstract We introduce a labelled corpus of stances about moral issues for the Brazilian Por- tuguese language, and present reference results for both the stance recognition and polarity classification tasks. The corpus is built from Twitter and further expanded with data elicited through crowd sourcing and labelled by their own authors. Put to- gether, the corpus and reference results are expected to be taken as a baseline for fur- ther studies in the field of stance recogni- tion and polarity classification from text. 1 Introduction Computational sentiment analysis may be under- stood as a wide range of tasks intended to iden- tify opinions, emotions and other types of stance expressed in natural language text (Tsytsarau and Palpanas, 2012; Liu, 2015). Among these, opin- ion mining is arguably the most well-studied form of sentiment analysis, consisting of identifying the target of an opinion, and/or the polarity (posi- tive, negative, neutral etc.) of the sentiment ex- pressed towards this target (Tsytsarau and Pal- panas, 2012). Stance recognition (Anand et al., 2011; Hasan and Ng, 2013; Lai et al., 2016; Mohammad et al., 2016b; Zarrella and Marsh, 2016; Wei et al., 2016; Mohammad et al., 2017), by contrast, consists of deciding whether the author of a piece of text shows a favourable or unfavourable attitude (or position) towards a certain target (Mohammad et al., 2017). The distinction between sentiment and stance is motivated by the observation that a sentiment, regardless of being positive or nega- tive, may reflect a favourable or unfavourable posi- tion towards the target (Mohammad et al., 2016b). For instance, given the target topic ’veganism’, a sentence as in ‘beef tastes wonderful’ expresses a positive feeling (which would indeed be recog- nised as such by traditional sentiment analysis sys- tems), but it also reflects an unfavourable position towards this particular target. Stance recognition from text is a well-known and yet challenging research topic. Systems of this kind enable the development of more complex sentiment analysis applications, and have been at the centre of a recent shared task (Mohammad et al., 2016b) focused on the use of supervised and unsupervised methods for stance recognition in the English language. For other less-resourced languages, however, resources remain scarce. Based on these observations, this paper presents a labelled corpus of stances in Brazilian Portuguese, and a number of computational models addressing two related issues: stance recognition, is presently regarded as the binary classification problem of deciding whether a given text conveys any attitude towards a certain target topic or not, and stance polarity classification, which is regarded as the binary classification problem of deciding whether a given stance expressed as text shows a positive or negative attitude towards the target topic. Examples of both …

Content truncated to 500 words