Semantic distances between comments in online newspaper forums increase over time

Abstract

[EN] In this longitudinal study, we analyze 75 million comments from the Austrian newspaper DerStandard's online forums between 2013 and 2022, focusing on the semantic dissimilarity of user comments over time. Using a specialized multilingual embedding model, we compute cosine distances between comment texts to track how textual expressions evolve. Our key finding reveals a strong trend of increasing dissimilarity between comments across the decade, independent of the total number of comments per month. Notably, this trend varies across different article topics, with health-related comments showing the highest increase in dissimilarity and lifestyle comments the lowest. We provide insights into the changing nature of online discourse, suggesting potential shifts in communication patterns, possibly influenced by significant events like the COVID-19 pandemic. Our study highlights the potential of text embedding techniques to capture nuanced changes in digital communication over time.