Post analysis

Detailed Post Analysis

A deeper dive into the semantic analyses of the 1,000 posts, including thematic analyses, post perspectives, and organizations or names mentioned in the posts.

Social Media & Phones at the Core of the Issue

Dimensionality reduction was conducted via UMAP, then posts were clustered with HDBSCAN in order to identify key topics discussed in these top 1000 posts. An overwhelming 4.6% of posts directly mention either social media, phones, or both. Posts include thoughts such as "All these events have been fully led by how addicted the world is to Social Media." and "Leave your phone at home and you will see how lonely the world really is right now." Other topics discussed include AI or generated content, as well as the internet more broadly.

Clusters shaded pink are explicitly about phones and/or social media, by label and top terms. Click any bar to see two example post titles from that cluster.

Redditors Discuss Psychological Effects, Impacts of Addiction, and the Social Necessity of the Digital World

Further thematic anlyses was condicted by using k-sparse autoencoder (k-SAE) on post embeddings, and an LLM was then called to identify semantic themes which rose from each latent's top-activating posts. Click one to see example posts from each theme identified.

Redditors Talk about Themselves

Computing the distribution of posts in first, second, and third-person point of view to determine whether posts are typically about themselves or other people. From the analysis, it is clear that a large majority of posts are about themselves, since 63% used first-person languate (ie: I and we)

Who are Redditors Mentioning in Posts?

Named-entity recognition was conducted over every post to identify people and organizations mentioned in the posts. 585 posts name at least one organization amd 133 posts name at least one person.

The bar chart below shows the top 20 most frequently mentioned organizations and people. To find results about a specific name, you may use the search bar above the chart.

Name variants of the same entity are combined (e.g. "Instagram" + "IG" + "insta"; "Elon Musk" + "Musk" + "Elon") — only high-confidence merges, so a bare first name like a standalone "Jacob" was left alone rather than guessed at.

Posts are Negative Across All Themes

A sentiment analysis was conducted on posts to determine the distribution of negative, neutral, and positive posts in each theme. Across all themes, negative posts take up the majority of the posts, exceeding 50% of posts in many themes as well.

0%50%100%
Negative Neutral Positive