Each comment was classified into one of six main categories: Neutral, Disagreement/Rejection,
Harassment, Insult (Offensive Language and Profanity), Discrimination, and Violence, with
subcategory breakdowns to capture the more specific nature of harmful speech (Figure 3).
Figure (3). Six main categories for annotation
The dataset consists of 33,716 social media comments collected during the election period,
analysed across two languages: Arabic (16,238 comments, 48.2%) and Kurdish (17,478 comments,
51.8%) see Table 6. The comments were attached to and taken from posts published by monitored
accounts across multiple platforms. In Arabic, 76 accounts were monitored — 47 on Facebook, 23 on
TikTok, and 6 on X (formerly Twitter) — spanning political figures, media outlets, political parties,
public pages, and influencers. In Kurdish, 84 accounts were tracked — 56 on Facebook and 28 on
TikTok — with the presence of female candidates on social media more noticeable (23 Facebook
accounts), in addition to public pages, and media outlets.
main
Disagreement/Rejection
Discrimination
Harassment
Insult (Offensive language and profanity)
Neutral
Violence
Total
Arabic
3297
1561
3001
1723
6463
193
16238
Kurdish
4690
2974
756
3918
5018
122
17478
Total
7987
4535
3757
5641
11481
315
33716
Table 6: Data sample analysis breakdown of Arabic and Kurdish social media comments
29