◆  Revisiting: analyzing Twitter's trending topics  ◆

In September 2021 I wrote Analisando os trending topics do Twitter - a small experiment that used Twitter's API and networkx to build a graph of hashtag co-occurrences. Each trending topic became a node; two hashtags appearing in the same tweet became an edge. The fun finding at the time: "tomate" (tomato) was bridging multiple unrelated discussion communities for reasons I never fully understood.

Reading it now, almost everything about the plumbing is dead. Almost everything about the method still works.

What broke

The Twitter API as it existed in 2021 effectively ended in 2023. The free tier was killed; the basic paid tier became expensive enough that hobbyist scripts like the one in that post are flatly impractical now. The search/tweets endpoint that powered the whole analysis is gated behind enterprise pricing.

So the code in the original post will not run today, and there is no point trying to fix it where it sits.

What survived

The actual analytical move - take a stream of public posts, extract co-occurring tags, build a weighted graph, look at the communities - is platform-agnostic. The platform died; the method outlived it. Modern places to point the same lens:

• Bluesky via the AT Protocol firehose. Free, open, fully public, no rate-limit dystopia. Probably the closest to "Twitter 2021" the social web has right now.
• Mastodon via the public timeline of a relay or instance. Federated, decentralised, smaller volume but cleaner signal.
• Reddit via the official API or pushshift's successors. Subreddits as natural communities; comment graphs as the edge structure.
• HackerNews via the Firebase API. Smaller corpus, but completely free and dense with technical hashtag analogues (story titles, tags, domains).

The interesting bit was never "Twitter." It was "what does the co-occurrence graph of public conversation actually look like, and what surprises you when you flatten it onto a graph instead of a feed?"

Where it would live today

If anyone wanted to do this in 2026, Bluesky is where it goes. The firehose is honest infrastructure - full stream, no per-query pricing, documented schema. The Twitter version of the open social web died. The Bluesky version is alive, and the kind of small experimental script that fueled the 2021 post is still allowed to exist there. That's mostly what I wanted to flag.