Stack Overflow data/ML question intelligence (weekly)

Every distinct Stack Overflow question surfaced by per-tag sweeps of a curated ~38-tag data/ML universe over the trailing 7 complete UTC days, deduplicated on question_id (first sweep wins as primary_tag; all observed tags unioned into tags). The official keyless Stack Exchange /questions API (sort=creation, fixed fromdate/todate window) supplies the raw material; the transform derives day-granular ages from the as-of stamp, scores each question with a documented 0-100 attention composite (60% min-max-normalized log10(1+views_per_day) + 40% min-max-normalized non-negative score_per_day), ranks it as attention_rank (ties: views desc, question_id asc), and flags unanswered_pain_flag when a question has zero answers with at least 2x the snapshot's median view count (visible community demand, no resolution). Titles + metadata only (CC BY-SA 4.0, attribution required) — no bodies, comments, or user data. Primary key: (week, question_id). Cadence: weekly. Nullability: scores and ranks are never null; age is at least 1 day. Caveats: counts tick intraday, so same-day re-runs refresh values; attention scores are within-snapshot relative (recalibrated every week). Sample use: order by attention_rank for the week's hottest data/ML questions, or filter unanswered_pain_flag for under-served demand.

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