Team Ned Leeds at SemEval-2019 Task 4: Exploring Language Indicators of Hyperpartisan Reporting

Proceedings of the 13th International Workshop on Semantic Evaluation (SemEval-2019), pages 1026–1031 Minneapolis, Minnesota, USA, June 6–7, 2019. ©2019 Association for Computational Linguistics 1026 Team Ned Leeds at SemEval-2019 Task 4: Exploring Language Indicators of Hyperpartisan Reporting Bozhidar Stevanoski, Sonja Gievska Faculty of Computer Science and Engineering Ss. Cyril and Methodius University Rugjer Boshkovikj 16, Skopje, Republic of North Macedonia bozidar.stevanoski@students.finki.ukim.mk sonja.gievska@finki.ukim.mk Abstract This paper reports an experiment carried out to investigate the relevance of several syntac- tic, stylistic and pragmatic features on the task of distinguishing between mainstream and par- tisan news articles. The results of the eval- uation of different feature sets and the extent to which various feature categories could af- fect the performance metrics are discussed and compared. Among different combinations of features and classifiers, Random Forest classi- fier using vector representations of the head- line and the text of the report, with the inclu- sion of 8 readability scores and few stylistic features yielded best result, ranking our team at the 9th place at the SemEval 2019 Hyper- partisan News Detection challenge. 1 Introduction Current influential technological megatrends, such as, smart phones and social networking often come with some unwanted side effects - prolific spread of false, biased and misleading informa- tion, for instance. In the past few years, the po- tential threats and consequences of disseminating fake news has reinforced the discussion on the re- sponsibility of social media and governments to tackle the issue, sooner rather than later. Enabling users to report on and be informed of untruth- ful, deceitful and fraudulent content and sources is expected to become a type of guiding principle for those involved in publishing and disseminating content online. There is a blurred line between deceptive writ- ing and hyperpartisan reporting, producing ex- tremely biased articles in favor of one political party, cause or individual, while preserving the format and appearance of professional articles. Adherence to the ethics and rules of objective re- porting is frequently debatable when it comes to political analysis in media articles. While certain truthful facts are present, they are carefully en- tangled in a narrative package with biased views, populistic messages and divisive topics, using lan- guage that polarizes and flares emotions. Rather than labeling and grading news articles on the truth continuum, researchers usually opt for identifica- tion of the phenomenological and contextual fea- tures of distinguishing hyperpartisanship in online news articles. People use diverse set of cues extrapolated from published text and external knowledge and sources, when verifying the veracity of informa- tion imparted by others. A large body of evidence documents the impact of deception has on lan- guage choices people make. A notable body of work exists revealing insights into the language of deceit in interrogation context (Porter and Yuille, 1996), court hearings (Coulthard et al., 2016), or personal relationships (Miller et al., 1986). Em- pirical studies still remain the primary manner in which manifestation of deceptive human behavior online is studied. Analysis of political language (Rashkin et al., 2017), …

Content truncated to 500 words