70k human Wikipedia navigation traces show collective knowledge escapes graph structure predictions
burny_tech · x · 2026-11-10
An arXiv paper analyzes 70,770 human navigation trajectories through Wikipedia via the Wikispeedia game, asking what distinguishes human collective navigation from graph structure and language statistics alone.
- Flow Trees merge all trajectories entering/leaving an article into group-split hierarchies; built for the 815 most-traversed articles.
- Human trees are deeper with longer trunks and later splits than four random-walker null models (unweighted, graph-weighted, word-frequency, semantic-similarity).
- Article degree explains 51.3-70.9% of branching-factor variation for random walkers but only 24.1-36.2% for humans.
- Authors define four article types — beacons, funnels, forks, corridors — from the dimensions graph structure fails to predict.
More from Research
- RIDArena: A New Benchmark Setting for Evaluating Rogue Internal AI Deployments — jammastergirish · 2026-11-10
- Open-source TotalSegmentator traces airways in 38s on one GPU, navigates to lung nodule — MaziyarPanahi · 2026-11-10
- EEG vs implanted BCIs: researchers clash over physical limits and AI decoding potential — burny_tech · 2026-11-10
- Charlotte Stix traces the history of "trustworthy AI" as a policy concept — ArtificialOther · 2026-11-10
- The Softmax Wall: in FP8/FP4 attention is no longer bottlenecked by matmuls — burny_tech · 2026-11-10
- ExploitGym wins Best Benchmark at Open Source AI Awards, flags OpenAI–Hugging Face incident — dawnsongtweets · 2026-11-10