Skip to content
tt

Discovering hot nodes

Walk the public graph from seeds and rank the hottest users, videos, hashtags, and sounds it reaches.

The single commands answer "tell me about this one thing." discover answers "start here and show me what is hot nearby." It walks the public graph outward from one or more seeds, scores every node it reaches, and emits a ranked, uniform row per node.

tt discover --seed-video https://www.tiktok.com/@tiktok/video/7106594312292453675 -o table

Each row is one node:

KIND   ID                   NAME              AUTHOR  METRIC    SCORE   DEPTH  VIA       URL
video  7106594312292453675  how many frogs..  tiktok  562500    0.4653  0      seed      https://...
user   MS4wLjABAA...         TikTok            tiktok  94400000  0.8861  1      authored  https://www.tiktok.com/@tiktok
sound  7106594280055130923  original sound    TikTok  0         0.4     1      uses_sound

KIND is one of user, video, hashtag, sound. METRIC is that kind's headline number (followers, plays, views, uses). SCORE is a 0..1 hotness rank. DEPTH is hops from a seed, and VIA is the edge the walk took to reach it. Every row keeps a url, so the full record is one tt video/tt user/... away.

Seeds

Start from anything. Seeds combine, so a single walk can fan out from several roots at once:

tt discover \
  --seed @tiktok --seed @nasa \
  --seed-tag minecraft \
  --seed-video 7106594312292453675 \
  --seed-sound 7106594280055130923 \
  --seed-search "frog" \
  --trending
Flag Seeds from
--seed @handle a user (repeatable)
--seed-tag name a hashtag (repeatable)
--seed-sound id a sound, by id or /music/ url
--seed-video id one video, by id or url
--seed-search "phrase" a search result page
--trending the trending feed

At least one seed is required.

How far it walks

--depth caps the hops from a seed (default 2). The walk is best-first: it always expands the hottest unexpanded node next, so a shallow run still surfaces the strongest nodes first. Bounds keep it honest:

tt discover --seed @tiktok --depth 3 --fanout 20 --max-nodes 200 --max-requests 800
  • --fanout is how many neighbors it takes from each list-bearing node.
  • --max-nodes stops after that many nodes are emitted.
  • --max-requests stops after that many fetches.
  • --comment-mine N additionally pulls up to N commenters per video as user nodes (off by default, and it rides the API plane).

Ranking and filtering

The rows arrive hottest-first. --top N keeps only the N highest-scored, and --min-score drops weak nodes:

tt discover --trending --top 25 --min-score 0.5 -o table

--kind restricts what gets emitted without changing what the walk explores, so you can crawl through videos but print only the users it found:

tt discover --seed-tag minecraft --kind user --top 50 -o jsonl

The edges

Add --edges path.jsonl to record the graph the walk traversed, one edge per line, alongside the node output:

tt discover --seed-video 7106594312292453675 --edges edges.jsonl -o jsonl

{"from_id":"7106594312292453675","from_kind":"video","to_id":"MS4w...","to_kind":"user","type":"authored"}
{"from_id":"7106594312292453675","from_kind":"video","to_id":"7106594280055130923","to_kind":"sound","type":"uses_sound"}
{"from_id":"7106594312292453675","from_kind":"video","to_id":"name:Minecraft","to_kind":"hashtag","type":"tagged"}
{"from_id":"7106594312292453675","from_kind":"video","to_id":"handle:Gorillo","to_kind":"user","type":"mentions"}

Feed both streams into a graph tool, or join them later by id.

What a datacenter IP can and cannot reach

This matters more for discover than for any single command, because the walk chains many surfaces together. Two of them carry the weight:

  • The page blobs (a video's own JSON, a profile's own JSON) need no signing and answer from anywhere. They give the seed video, its author, its mentioned users, its sound, and its hashtags.
  • The list endpoints (a user's posts, a video's related feed, a hashtag's or sound's videos, search) ride the signed API plane. TikTok's firewall gates that plane from datacenter IPs.

So from a datacenter IP a --seed-video walk still reaches the small constellation around that video through the page blob, but it cannot page a feed to go wider. The walk runs correctly and says exactly what it could not reach. The stderr summary names the walled surfaces:

reached 4 node(s) (sound 1, user 2, video 1), 4 edge(s), 6 request(s); 1 walled (related 1); stopped: frontier drained

When the wall stops the walk cold (every seed needs the API plane and nothing came back), discover exits 4. From a residential session the list endpoints answer and the walk goes as wide as the bounds allow. See troubleshooting for the full surface map.

4 min read