Development of methods for extracting situations, emotional assessments and values from social media texts

Abstract

This paper presents an end-to-end pipeline for automated social media discourse analysis, integrating situation extraction from natural language texts, emotion classification, and value identification. Most existing approaches are limited to binary sentiment analysis and do not account for the value context underlying users’ emotional responses; this work addresses that gap. The pipeline consists of four sequential stages: data collection from VKontakte via API, generation of situational scenarios per comment using Schoemaker’s scenario planning framework, emotion classification using a fine-tuned RuBERT model, and value identification according to Schwartz’s theory using the DeepSeek large language model. The novelty lies in the joint analysis of emotions and values within a unified pipeline applied to Russian-language social media – an approach not previously implemented for this language and platform. A corpus of 2,721 situations was constructed from 950 real VKontakte comments across five thematic domains. The emotion classifier achieved Macro F1 = 0.798, outperforming the best TF-IDF baseline by 10.2 percentage points. Value annotation quality reached 90.8 % accuracy with Cohen’s Kappa κ = 0.863. Results reveal the dominance of Security (34.1 %) and Universalism (19.2 %) values in the Russian-language VKontakte segment, confirming the pipeline’s applicability for monitoring sociocultural transformations in digital environments.