Public Perceptions of Telegram’s Association with Illicit Activities: A Sentiment Analysis Using VADER and Machine Learning

Abstract

Digital communication platforms have significantly transformed social interaction and information dissemination, yet simultaneously present challenges related to illicit activities, security threats, and regulatory oversight. Telegram, a widely-used encrypted messaging service, has recently drawn global scrutiny due to allegations linking it to criminal enterprises, including identity theft, illicit drug markets, and distribution of child exploitation materials. This study systematically evaluates public sentiment surrounding Telegram’s reported facilitation of illegal activities, employing comparative sentiment analysis methodologies: VADER (Valence Aware Dictionary and Sentiment Reasoner) and a supervised machine learning approach (TF-IDF vectorization coupled with Logistic Regression). A corpus of 632 reader comments from a Wall Street Journal article discussing Telegram’s controversial associations was analysed. VADER-based labelling identified an unexpectedly predominant positive sentiment (59.5%), indicating potential public scepticism toward negative media narratives or ideological support for encrypted platforms. The logistic regression classifier demonstrated robust predictive performance, with overall accuracy of 89.56%, precision of 91%, recall of 90%, and an F1-score of 89%, yet displayed a notable positivity bias, misclassifying nuanced negative commentary. Qualitative word cloud visualizations further highlighted distinctive lexical patterns, underscoring explicit concerns around security and criminality in negative comments and humour or reflective discourse in positive remarks. Methodologically, results expose critical limitations of traditional lexical approaches in capturing subtle, implicit, or context-dependent negativity, suggesting the integration of advanced context-aware modelling techniques, such as transformer-based neural embeddings, for enhanced precision. Practically, this analysis provides critical insights for platform governance, risk management strategies, regulatory frameworks, journalistic practices, and computational linguistics research, emphasizing the necessity for balanced methodological approaches to accurately gauge and respond to nuanced public sentiment within contentious digital discourse contexts.