TIMING
Wikipedia pageview forecasts, no LLM, just stats
Start new thread

TIMING - Wikipedia pageview forecasts, no LLM, just stats

by•
For any Wikipedia topic, TIMING pulls 90 days of daily pageviews from the Wikimedia Analytics API, computes a 7-day rolling average and a linear regression on log(1+views), and flags rupture points with a z-score at 2σ (after subtracting the expected false-positive rate from pure noise). It outputs two normalized scenarios (continuity vs rupture), and every sentence is a template filled from the actual computed numbers, no LLM involved. Free, no account needed.

Add a comment

Replies

Best
Maker
📌
Hey! Maker here. I wanted a forecasting tool where you can see every step, so TIMING uses no language model. For any Wikipedia topic it takes the last 90 days of daily pageviews (Wikimedia Analytics API), works on log(1+views), and computes a 7-day rolling average, a linear regression, and rupture points with a z-score at 2σ. One detail I like: on pure noise, a 2σ threshold flags about 4.6% of points by chance, so the rupture score first subtracts that expected false-positive rate before extrapolating. The two scenarios (continuity vs rupture) are normalized to 100%, and every sentence is a template filled with the numbers shown next to it. Important caveat: this models public attention, not events. Data sources: Wikimedia pageviews, Wikipedia, Wikidata, OpenStreetMap. No account needed.