Are We Playing Wikispeedia All Wrong?
Mixed Linear Model over 76k+ Wikispeedia gameplay paths (n=50,138) identifies semantic-similarity navigation as the dominant strategy: β=−21.3 s, p<0.001.
What this is
Study of 76k+ Wikispeedia gameplay paths to find which navigation strategy actually shortens completion time. Players take 2.4× longer than the shortest path on average. Plenty of room to study which heuristics help.
How it works
Data: 51k completed paths, 25k abandoned. Tested four strategies:
- Hub-based: route through high-PageRank pages.
- Semantic: click the link closest to the target in TF-IDF space.
- Top-link: click links high in the page layout.
- Backtracking: revisit earlier pages.
Fit a Mixed Linear Model with backward selection (n=50,138). Per-game difficulty as a random effect, each strategy’s contribution to completion time as the fixed effects.
Results
Semantic navigation wins: β = −21.3 s per unit semantic-increase score (p<0.001). Top-link helps too (β = −8.2 s). Backtracking hurts (β = +37.5 s). Hub usage barely moves the needle (β = +2.4 s).
CS-401 Applied Data Analysis at EPFL, team “5pack”. Full write-up at the linked site.