App Store review-aspect sentiment (agent-curated)
Daily enrichment panel scoring customer-review sentiment for 32 prominent US App Store apps (social, finance, shopping, streaming, productivity, navigation, food delivery, travel). Each run verifies the roster against Apple's keyless lookup API by bundleId, fetches up to 4 pages of the most recent customer reviews per app (keyless customerreviews RSS JSON), and ships trailing-30-day aggregates: review volume, rating distribution (avg, 1-2-star and 5-star shares), lexicon sentiment (review_sentiment = 100 * tanh(S/4) over a documented ~70-word valence lexicon with negation handling, S = signed weight sum), per-aspect sentiment for a documented 8-aspect keyword taxonomy (bugs, pricing, features, ux, performance, support, ads, privacy; null with < 5 mentions), health_score = 100 * (0.45 * norm(avg_rating) + 0.35 * norm(mean_sentiment) + 0.20 * (1 - clipped 1-2-star share)) with a deterministic health_rank, version_dip_flag when the latest version's 1-2-star share exceeds the rest-of-window share by >= 15 pp (>= 10 reviews on each side), and the worst qualifying top_aspect. Who joins this: a product/brand team joins daily per-app review-aspect sentiment to their release/QA calendar on bundle_id + as_of to catch launch regressions, and a competitive-intelligence team benchmarks apps within a category on category + as_of (primary key (as_of, bundle_id); join keys bundle_id, app_name, as_of, category). Caveats: aspect assignment is keyword-based, not semantic; sentiment is lexicon-based (sarcasm/idiom escape it); apps with < 10 in-window reviews keep null-scored rows flagged data_sufficient = 0; the mostRecent feed only exposes the newest ~200 reviews, so low-velocity apps have thinner windows. No review text, titles, or author names are stored — only derived numeric aggregates.
- signals
- technology
- daily
- sentiment