The ‘Sessiz İstila’ Dataset: Emotion-Annotated Turkish Tweets on Anti-Refugee Discourse and Emoji Lexicon

This article describes an emotion-annotated corpus of Turkish-language posts published on X (formerly Twitter) that discuss ‘sessiz istila’ (‘silent invasion’), a contested phrase used in Turkish public debate to frame the presence of refugees, particularly Syrian refugees, as a threat. We retrieved posts using the academictwitteR interface to the Twitter Academic Research API with the keyword ‘sessiz istila’, restricted to publicly available Turkish-language posts published between 1 June 2021 and 31 December 2022, yielding 47,024 posts. We normalized every post with a rule set implemented in R that replaces retweet markers, URLs, user mentions, and hashtags with Turkish placeholder tokens, converts text to lowercase, and substitutes each emoji with a Turkish description drawn from a purpose-built emoji lexicon of 3,664 entries released alongside the corpus. Emotion labels were assigned automatically by an Emotion Recognition Model obtained by fine-tuning BERTurk (bert-base-turkish-cased) on a class-balanced subset of the TREMO dataset containing 18,018 validated entries, 3,003 for each of the six Ekman categories (happiness, fear, sadness, disgust, surprise and anger). We retained predictions with softmax confidence below 0.6 but flagged them as ‘ambiguous’ rather than discarding them so that users can apply their own confidence policy. The released corpus therefore contains 40,880 confidently labeled posts and 6,144 ambiguous posts, each with its creation timestamp, normalized text, date, year, and emotion label. The dataset supports secondary research on online xenophobia and anti-immigrant sentiment, on event-driven emotional dynamics in social media, and on benchmarking and error analysis of emotion classifiers for Turkish, an agglutinative and comparatively low-resource language. The balanced TREMO training file and the Turkish emoji lexicon are reusable across corpora.