GDELT global-news company tone (daily)

Daily enrichment signals scoring how the world's news talks about ~70 US and Canadian companies. Each day the 96 keyless GDELT 2.1 GKG fifteen-minute bulk files for the last complete UTC day are downloaded; articles are deduplicated on URL, their V2ENHANCEDORGANIZATIONS are resolved against a curated company dictionary (name, primary-listing ticker, HQ country, aliases — shared with the hn_company_sentiment series so curation cannot drift), and GDELT's own per-article V1.5TONE scores (computed from full article text) are aggregated per company and day. Per company: n_articles, n_sources, mean/median/std/min/max tone, pos_share / neg_share (fraction of articles with tone >= 1.0 / <= -1.0), mean_pos_words_pct / mean_neg_words_pct, dominant_theme, tone_signal (positive / negative / polarized / none), and news_tone_score = 100 * (55% min-max-normalized mean_tone + 25% min-max-normalized pos_share + 20% min-max-normalized (1 - neg_share)), ranked as tone_rank (1 = most positive; ties: n_articles desc, company name asc). Who joins this: a quant desk joins per-ticker daily global-news tone to their positions/returns table on ticker + as_of; an IR/comms team joins on company_name to their media-monitoring calendar. Primary key: (as_of, company_name); join keys: ticker, company_name, as_of, country_code (ISO alpha-3). Nullability: ticker is null for private companies (OpenAI, Anthropic, Stripe, ...); std_tone is null for single-article companies; dominant_theme is null when no themes were recorded. Caveats: news_tone_score is a within-snapshot relative measure, not comparable across days; GDELT tone is computed on machine-translated text for non-English articles; organization NER misses and corporate-suffix collisions are possible despite suffix stripping. No article text and no per-row URLs are stored. Sample use: filter ticker = 'TSLA' for Tesla's daily global-news tone, or order by tone_rank for today's most positively covered companies.