Text embeddings and clustering for characterizing online communities on Reddit

This work analyses Reddit, the largest public, topiccentered social forum.In the experiments, contextualized text embeddings, obtained using DistilBERT, represented subreddit content.Next, clustering was performed, using an unsupervised K-means algorithm and evaluated with multiple clustering metrics.The obtained clusters were analyzed.Moreover, changes of cluster structure, between 2019 and 2022 have been examined.