Mapping offensive discourse in sports on X platform: a weakly supervised NLP and journalism-based analysis

Creative Commons License

Shbair A., Haqmal L. U. R., Livberber T., Günay M.

ONLINE MEDIA AND GLOBAL COMMUNICATION, sa.1, ss.1-25, 2026 (Scopus)

Özet

Purpose: Social media platforms have become a central place for sport-related

public discourse that often involves emotionally charged and offensive language.

This study examines offensive sports discourse on X (formerly Twitter) through an

interdisciplinary approach that combines natural language processing (NLP) and

journalism research.

Design/methodology/approach: A dataset of 7,279 sport-related tweets that were

collected between January and December 2025 was analyzed. An initial set of 350

tweets were manually annotated first and then used to train a machine learning

classifier based on TF-IDF features and logistic regression. The classifier got an

accuracy of 0.71. A confidence-based weakly supervised labelling strategy was then

applied, resulting in 2,042 high-confidence labelled tweets.

Findings: Quantitative analysis shows that offensive tweets demonstrated higher

average engagement levels than regular tweets in terms of likes, retweets, views, and

replies, although these differences were not statistically significant. Unverified users

predominantly produce offensive content and this contents appear more frequently

in original tweets than in replies. Lexical analysis reveals that offensive discourse

combines explicit profanity with sports-related terminology. The findings demon-

strate how NLP-based methods can support journalism research by revealing large scale patterns in digital sports communication, and how these methods can be used

in interdisciplinary settings to provide unique and novel insights through the

analysis of the available data.

Practical implications: This research assists the growing field of computational

journalism by leveraging NLP techniques with journalism-based research in an

interdisciplinary settings to gain unique and novel insights about the domain-

specific online sports discourse.

Social implications: This study helps us to achieve empirical insights and intuitions

into fan behavior, interaction and the pattern of their engagement. At the same time,

it is very important to perform journalistic interpretation for the purpose of

contextualizing computational findings within ethical, cultural, and professional

frameworks.

Originality/value: This study adopts established NLP based methods to explore

large scale communicative patterns in online sports discourse.