UPV-28-UNITO at SemEval-2019 Task 7: Exploiting Post’s Nesting and Syntax Information for Rumor Stance Classification

Proceedings of the 13th International Workshop on Semantic Evaluation (SemEval-2019), pages 1125–1131 Minneapolis, Minnesota, USA, June 6–7, 2019. ©2019 Association for Computational Linguistics 1125 UPV-28-UNITO at SemEval-2019 Task 7: Exploiting Post’s Nesting and Syntax Information for Rumor Stance Classification Bilal Ghanem1, Alessandra Teresa Cignarella1,2, Cristina Bosco2, Paolo Rosso1, Francisco Rangel1,3 1. PRHLT Research Center, Universitat Polit`ecnica de Val`encia 2. Dipartimento di Informatica, Universit`a degli Studi di Torino 3. Autoritas Consulting bigha@doctor.upv.es, cigna@di.unito.it, prosso@dsic.upv.es, bosco@di.unito.it, francisco.rangel@autoritas.es Abstract In the present paper we describe the UPV- 28-UNITO system’s submission to the Ru- morEval 2019 shared task. The approach we applied for addressing both the subtasks of the contest exploits both classical machine learn- ing algorithms and word embeddings, and it is based on diverse groups of features: stylistic, lexical, emotional, sentiment, meta-structural and Twitter-based. A novel set of features that take advantage of the syntactic information in texts is moreover introduced in the paper. 1 Introduction The problem of rumor detection lately is attracting considerable attention, also considering the very fast diffusion of information that features social media platforms. In particular rumors are facili- tated by large users’ communities, where also ex- pert journalists are unable to keep up with the huge volume of online generated information and to de- cide whether a news is a hoax (Procter et al., 2013; Webb et al., 2016; Zubiaga et al., 2018). Rumour stance classification is the task that in- tends to classify the type of contribution to the rumours expressed by different posts of a same thread (Qazvinian et al., 2011) according to a set of given categories: supporting, denying, query- ing or simply commenting on the rumour. For in- stance, referring to Twitter, once a tweet that intro- duces a rumour is detected (the “source tweet”), all the tweets having a reply relationship with it, (i.e. being part of the same thread), are collected to be classified. Our participation to this task is mainly focused on the investigation of linguistic features of social media language that can be used as cues for de- tecting rumors1. 2 Related work The RumorEval 2019 shared task involves two tasks: Task A (rumour stance classification) and Task B (verification). Stance Detection (SD) consists in automatically determining whether the author of a text is in favour, against, or neutral towards a given tar- get, i.e. statement, event, person or organiza- tion, and it is generally indicated as TARGET- SPECIFIC STANCE CLASSIFICATION (Mohammad et al., 2016). Another type of stance classification, more general-purpose, is the OPEN STANCE CLASSIFI- CATION task, usually indicated with the acronym SDQC, by referring to the four categories ex- ploited for indicating the attitude of a message with respect to the rumour: Support (S), Deny (D), Query (Q) and Comment (C) (Aker et al., 2017). Target-specific stance classification is especially suitable for analyses about a specific product or political actor, being the target given as already extracted, e.g. from conversational cues. On this regard several shared tasks have been organized in recent years: see for …

Content truncated to 500 words