Lithuanian Emotional Manipulation Dataset

This dataset contains comments extracted from Lithuanian news portals, along with fragments (spans) of different techniques of emotional manipulation language identified in these comments. The spans were identified using a GPT-4.1 large language model and four prompt engineering strategies. Each prompt strategy uses different prompts from the dataset. The dataset contains both raw data returned by the model and pre-processed data with aligned spans and their positions within a comment. It also includes the gold dataset with manually annotated spans of emotionally manipulative language.

Content Notice: This dataset contains authentic user-generated comments that may include offensive, non-normative, or inappropriate language, such as profanity, insults, and emotionally manipulative expressions. The content is provided strictly for research and analytical purposes.

Some material may be disturbing or unsuitable for certain audiences.

Licence Notice: This dataset includes LITIS corpus comments, licensed under CLARIN-LT Public End-User Licence (PUB).