Choose Social Media Datasets You Can Explain and Share

Start with the question your public table should answer. Comparing post performance needs one row per post. Studying comment labels needs one row per comment—or a safer summary by source. Exploring a friendship network needs one row per connection.

These three social media datasets cover those different tasks. For a first publishing project, Facebook Metrics is the most straightforward option: it is small, tabular, and has an explicit reuse license. The other two introduce text-handling and network-structure decisions.

1. Facebook Metrics: compare historical post performance

UCI’s Facebook Metrics dataset contains 500 posts published in 2014 on one cosmetics brand’s Facebook page. The release includes only part of the 790 rows and features used in the associated research; other material was omitted for confidentiality. UCI provides a CSV and lists fields including post type, month, weekday, hour, paid status, reach, impressions, and engaged users.

Useful project: compare the distribution of lifetime post reach across post types, with paid status available as a filter.

Suggested public row: one released post, retaining the original metrics and adding a documented release-specific row identifier if needed. Do not manufacture full timestamps from month, weekday, and hour fields.

Interpretation limit: this is a historical sample from one brand, not a benchmark for current Facebook performance. A difference between paid and unpaid posts does not by itself establish that payment caused the difference.

UCI lists the dataset under CC BY 4.0, making it a practical starting point for an attributed public table. Keep the original metric names or document any renaming beside the page.

2. YouTube Spam Collection: summarize labeled comments

UCI’s YouTube Spam Collection contains 1,956 real comments across five video-specific CSV files. Its fields include comment ID, author, date, content, and a binary classification label. The collection was assembled for spam research from five videos that were among the ten most viewed during the collection period. UCI lists it under CC BY 4.0.

Useful project: publish counts and the share of comments labeled spam for each source file.

Suggested public row: one source file or video, rather than one identifiable commenter. Retain the source filename before combining the CSVs, and verify the label encoding against the supplied documentation before calculating counts.

Interpretation limit: the spam share describes this research collection, not YouTube overall. The selected videos and class composition matter; the archive is not a random sample of the platform.

A license is not a reason to publish every field. For a count-based page, leave author names, comment IDs, raw comments, and embedded links out of the public export. CC BY 4.0 itself notes that privacy and other rights may still limit a use.

3. SNAP Facebook Social Circles: explore connections

Stanford SNAP’s Facebook Social Circles dataset contains anonymized networks collected from survey participants. Its combined network has 4,039 nodes and 88,234 edges. A node represents a user in the network, not necessarily a survey participant; an edge represents a connection.

Useful project: derive a table showing each anonymous node’s number of connections within the combined network.

Suggested working row: two node IDs representing an undirected connection. The file documentation explains that individual ego-network files omit the central user, whereas the dataset page’s combined-network statistics include those users and their connections. Choose the combined file if you want to reproduce the reported combined totals.

Interpretation limit: this is not a list of all Facebook users or friendships. Profile feature meanings were deliberately obscured; do not infer identities or restore those meanings.

SNAP’s dataset page describes downloads and anonymization but does not state an explicit reuse license. Treat it as an analysis source, not an automatically cleared republication source: confirm redistribution terms before uploading raw or derived data publicly.

Turn the chosen dataset into a readable public table

Define the output before cleaning

For the YouTube summary, use this proposed schema, not a claimed dataset result:

Column	Meaning
source_file	Original CSV filename; one row per file
spam_comments	Count of records labeled spam
other_comments	Count of records labeled non-spam
total_comments	Sum of the two counts
spam_share	Spam count divided by total count, stored as a decimal
retrieved_on	Date you downloaded this release

This structure gives readers a comparison without exposing comment-level details.

Check the exported file

• Preserve identifiers as text and test the intended row key for blanks and duplicates. See primary keys for published CSV datasets.
• Distinguish zero from missing or unknown values; do not silently replace blanks with zero.
• Keep counts numeric. Document whether a proportion is stored as 0.25 or 25.
• For the YouTube aggregate, reconcile totals to each source file and check that the combined input contains the expected 1,956 records. Investigate missing or unexpected labels rather than silently excluding them.
• Remove personal details and unnecessary text before export—not after upload.

Add provenance, then publish

Include the dataset title, creators, source link, license where established, collection period, retrieval date, and transformations. For CC BY 4.0 material, provide credit, link the license, and indicate changes. Keep retrieval date separate from collection date: downloading a historical archive today does not make it current data.

TablePage accepts CSV, TSV, XLSX, and XLS and generates a public dataset page with a filterable table. Upload only the reviewed publication file. Inspect the resulting page for correct columns, numeric values, and readable descriptions before sharing its link.

Name the page for its actual scope—such as “Spam Labels in Five Archived YouTube Comment Samples”—rather than implying a platform-wide finding.