This is the repository for the project on computation analysis of dehumanization of Ukrainians on Russian Telegram.

Dehumanization (“the act of perceiving or treating people as less than human”) is a process that heavily contributes to extreme intergroup bias. Analysis of dehumanization at scale can provide a fuller view of the social environment while direct assessment is not feasible. The comprehensive literature review for this project is available through the Researchgate. We aimed to adapt and apply advancements in dehumanization conceptualization to the Russian language and the context of the 2022 Russian invasion of Ukraine.

The dataset used in the analysis consists of posts from popular political and news Telegram blogs. The data was collected from 25 November 2022 to 25 December 2022. The earliest post is dated 22 September 2015 at 16:12:02 UTC. We have concluded experimentation by the 28 May 2023. The experiments involved various techniques, including augmentation and feature engineering.

To data is available upon the reasonable request. Our best SpERT model trained for span-based dehumanization detection is available here. Original implementation of SpERT is outlined in paper and available in this repository.