Wikipedia company-article attention velocity (daily)

Daily enrichment signals measuring public attention to ~79 US and Canadian companies through English-Wikipedia pageviews. For each company the trailing 60 complete UTC days of daily views are pulled from the official keyless Wikimedia pageviews API for the company's canonical article (resolved by hand against the curated company dictionary shared with hn_company_sentiment/gdelt_company_tone; "Crusoe" has no article and is excluded), then transformed: views_60d (identity-gated: sum of daily views), mean_daily_views, views_7d / views_7d_prev (trailing vs prior week), wow_pct (views_7d-views_7d_prev)/views_7d_prev (null when the prior week is zero and the trailing week is also zero; +2.0 clip bound when attention appears from nothing), z_7d = (mean_7d - mean(baseline)) / stdev(baseline) over the first 46 days (flat baseline: +3.0 if the trailing week is above it, -3.0 below, 0.0 equal), attention_score = 100*(0.50*min-max(log1p(views_7d)) + 0.30*min-max(clip(z_7d,-3,3)) + 0.20*min-max(clip(wow_pct,-2,2))), min-maxed within the snapshot, attention_rank (score desc; ties: views_7d desc, company_name asc), spike_flag = 1 when z_7d >= 2.0, source_url = the canonical article URL (the pageviews REST fetch URL's timestamp runs trip the hub PII screen, so the article URL is stored instead). Articles with < 57/60 days of coverage are dropped loudly. Who joins this: an IR/comms team joins daily public-attention velocity per ticker to their trading/media calendar on ticker + as_of; a quant desk joins attention_score/z_7d to returns on ticker + as_of. Primary key: (as_of, company_name); join keys: ticker, company_name, as_of, country_code (ISO alpha-3). Nullability: ticker null for private companies (OpenAI, Anthropic, Databricks, ...); wow_pct null only when both weeks are zero. Caveats: attention_score is within-snapshot relative, not comparable across days; pageviews count all access types including automated traffic not already filtered by the API; article renames are absorbed by the canonical-title table. Underlying aggregates are CC0 1.0, so commercial_use = yes. Sample use: filter ticker = 'NVDA' for Nvidia's daily attention velocity, or order by attention_rank for today's fastest-rising company attention.

Les titres et les descriptions proviennent des sources de données, en anglais.