What did de-anonymization research on social media (e.g. Twitter) datasets demonstrate?

That "anonymized" social datasets can be re-identified by cross-referencing posting patterns, timing, and social-graph structure against publicly available data from other platforms.

Researchers showed that stripping names is nowhere near enough. The attack vectors used:

• Temporal posting patterns — when you post is a fingerprint.
• Network topology analysis — the shape of your friend/follower graph is highly unique.
• Writing-style fingerprinting (stylometry) — how you write identifies you.
• Cross-platform correlation — matching the anonymized graph to a public one.

The lesson: social graphs and behavioral patterns are themselves powerful quasi-identifiers. Removing the obvious identifiers leaves a rich, unique behavioral signature behind.