EthnoHate2 ( Ethnic Hate Speech Prediction in Social Media Texts )

Project leader: Olessia Koltsova

Project participans: Anton Surkov

This research is an ideological and methodological successor to the project of 2020 year and utilizes the same dataset as the primary source of data.
The study is devoted to developing models for the automatic detection of ethnic conflict in informal texts with an ethnic orientation.

The task is solved within the paradigm of pre-training (fine-tuning) language models such as transformer-encoders for classification.
Additionally, the research explores the influence of various augmentations (including generating alternative representations of text by large language models - LLM) on the quality of the models.

Furthermore, the study investigates the possibilities of using large language models to extract relevant information from texts with an ethnic orientation, as well as the automatic annotation of these texts without pre-training.