Dynamics of Toxicity Spread in Social Networks: A Similarity-Based Network Approach

The rapid spread of toxic content on social media poses significant societal challenges. Regulating harmful content in social networks is essential to maintain social harmony. This study explores whether the spread of toxic language or sentiments can be analysed using a dataset that includes comments labelled as toxic or non-toxic but lacks user and temporal information. The proposed approach uses similarity-based networks like Cosine, Jaccard, and Soft cosine and the SEIZ epidemiological model to study toxicity dynamics. While the analysis is theoretical due to data limitations, the findings provide valuable insights for planning prevention strategies, designing policies, and understanding the intrinsic properties of data. The results show consistent patterns across different networks, highlighting the important role of skeptics in slowing toxicity spread. This study demonstrates that even with limited data, theoretical modelling using similarity networks and SEIZ can reveal meaningful insights into the spread of toxic behaviour.