<p>This study analyzes Mastodon, a prominent federated social network made of several interconnected instances, and investigates community detection <i>at the instance level</i> by integrating content-based edge weighting to enhance modularity maximization techniques. Traditional community detection methods rely on structural connectivity, but in highly interconnected and dense networks, such as the one obtained through our crawling phase on Mastodon instances, this approach often fails to reveal meaningful clusters. To overcome this limitation, we computed a weighted graph by designing a process pipeline that assigns semantic similarity scores to edges, derived from trending posts. The effectiveness of this content-aware approach was evaluated using the Louvain and Leiden algorithms, showing a notable increase in modularity and community granularity compared to the results obtained on the same, but unweighted, graph. This study contributes to the field of community detection in decentralized networks by exploring the use of content-aware edge weighting, which provides insights into community structures shaped by semantic relationships. The findings enhance the understanding of decentralized social networks and offer a basis for further research into the impact of content similarity on community dynamics.</p>