A Pipeline for Graph-Based Monitoring of the Changes in the Information Space of Russian Social Media during the Lockdown

TL;DR

This paper presents a data processing pipeline for analyzing weekly changes in Russian social media during COVID-19 lockdown, revealing social behavior shifts and information dynamics through topic similarity and user clustering.

Contribution

It introduces a novel pipeline combining topic similarity time series and user clustering on social media data during a pandemic, with a case study on Russian platforms.

Findings

Social processes and behaviors are reflected in social media dynamics.

The pipeline effectively traces adaptation trends during the pandemic.

Psychological and sociological trends are observable in the data.

Abstract

With the COVID-19 outbreak and the subsequent lockdown, social media became a vital communication tool. The sudden outburst of online activity influenced information spread and consumption patterns. It increases the relevance of studying the dynamics of social networks and developing data processing pipelines that allow a comprehensive analysis of social media data in the temporal dimension. This paper scopes the weekly dynamics of the information space represented by Russian social media (Twitter and Livejournal) during a critical period (massive COVID-19 outbreak and first governmental measures). The approach is twofold: a) build the time series of topic similarity indicators by identifying COVID-related topics in each week and measuring user contribution to the topic space, and b) cluster user activity and display user-topic relationships on graphs in a dashboard application. The paper describes the development of the pipeline, explains the choices made and provides a case study of the adaptation to virus control measures. The results confirm that social processes and behaviour in response to pandemic-triggered changes can be successfully traced in social media. Moreover, the adaptation trends revealed by psychological and sociological studies are reflected in our data and can be explored using the proposed method.