saturn

This dataset catalogs 143,258 movies from Rotten Tomatoes across 16 columns covering metadata (title, director, writer, distributor), release info, runtime, genre, language, ratings, and critic/audience scores. Coverage is highly uneven — fields like boxOffice (89.7% null), rating (90.2% null), tomatoMeter (76.4% null), and releaseDateTheaters (78.5% null) are sparse, while audienceScore is missing in roughly half the rows. Worth a closer look first: the genre distribution, which is dominated by Drama (27,860), Documentary (15,162), and Comedy (11,514), and runtimeMinutes, which is heavily right-skewed (skew 7.6, max 2,700 minutes) with ~11.4% flagged as outliers despite a tight IQR of 84–103 minutes. The tomatoMeter and audienceScore distributions also tell a clear story — critics skew positive (median 73) while audiences are more middling (median 57). English dominates originalLanguage at 65.7% of titles, so any language-based analysis will be lopsided.

Slug-style identifier column: every value is a single token (one_word_rate 1.0, word_mean 1.0) with mean length ~18 chars and 142052 uniques out of 143258 rows. The 1206 duplicates (0.84%) are surprising for an id field — top repeats like 'catch_me_if_you_can' and 'hear_no_evil' suggest these are title-derived slugs rather than guaranteed-unique keys. Readability score is meaningless here (−75.5) because the tokens are underscore-joined phrases, not prose.

Short titles (mean 17 chars, median 3 words) of what look like films or works — top values include 'The Return', 'A Christmas Carol', 'Hero', 'Blue'. Predominantly English (3946) but 29 other languages are detected, with Spanish (123), German (80), and French (72) most common. Notable duplication: 16,488 repeats (11.5% duplicate rate) across 126,403 unique values out of 143,258 rows, and 17% are single-word titles.

This is an audience rating score on a 0-100 scale with 101 unique integer values, mean 55.67 and median 57. The distribution is wide (std 24.55, IQR 39) and slightly left-skewed (skew -0.23, kurtosis -0.83) with no outliers flagged. The dominant concern is missingness: 48.87% of rows are null, so nearly half the dataset lacks this score.

This is the Rotten Tomatoes critic score (tomatoMeter), a 0-100 percentage with 101 unique integer values, mean 65.77 and median 73. The distribution is left-skewed (skew -0.65) with Q1 at 45 and Q3 at 89, indicating most rated titles lean favorable. The dominant concern is coverage: 76.35% of rows are null, so the field is only populated for a minority of records.

This is a content rating field mixing theatrical (R, PG-13, PG, NC-17, G) and television (TVPG, TV14, TVMA, TVY7, TVG) classifications across 10 distinct values. The column is 90.23% null, so only ~9.77% of the 143,258 rows carry a rating, and within those R alone accounts for 55.28% of values. The mixed rating systems and the long tail (TVG, TVY7, G each appearing once) suggest inconsistent sourcing rather than a clean controlled vocabulary.

This column stores content-rating descriptors (e.g. 'Language', 'Violence', 'Some Sexual Content') serialised as Python-style list literals rather than clean arrays. It is 90.23% null and, among the 14k populated rows, 40.3% are duplicates with only 8,353 unique values across 143,258 records. A handful of non-English entries (12 it, 5 ro, 1 km) appear despite the vocabulary being tiny (1,188 words), and the bracket/quote artefacts in top_words confirm the values were never parsed out of their string representation.

This is a theatrical release date stored as an ISO-format string (every value is exactly 10 characters and a single token, e.g. '2018-09-14'). It is sparsely populated — 78.52% null — and the non-null values are heavily repeated, with a 60.8% duplicate rate across 12,062 distinct dates. The 'allcaps' alert is a false positive driven by digits-only strings; there's no actual text content to mine.

This is a streaming-release date stored as ISO-8601 text (len_median 10, one_word_rate 1.0, all top values match YYYY-MM-DD). Roughly 44.56% of rows are null and the duplicate_rate is 0.94, with a single date 2017-05-22 appearing 1232 times — heavy clustering on a few release days. The text-style alerts (allcaps, one_word, short_text) are artifacts of the date format, not a quality issue.

Movie or episode runtime in minutes, with a typical feature-length distribution (median 92, IQR 84-103). The tail is extreme: max 2700, skew 7.62, kurtosis 598.65, and 11.37% of rows flagged as outliers, suggesting a mix of shorts, multi-part specials, or full series totals alongside standard films. Roughly 9.65% of rows are null.

This is a categorical genre label for films, often a single word like 'Drama' (27,860 rows) or 'Documentary' (15,162) but sometimes a comma-separated combo such as 'Comedy, Drama'. With only 66 distinct vocabulary tokens but 2,912 unique strings and a 97.8% duplicate rate, the cardinality comes entirely from how genres are concatenated. Note the 7.74% null rate and that 55% of values are single-word — multi-genre rows are the minority.

Categorical language label with 112 distinct values, dominated by English at 65.7% of non-null rows (85,034 of 143,258). The long tail spans regional variants (e.g., 'French (Canada)' vs 'French (France)', 'English (United Kingdom)') alongside bare language names like 'English' and 'French', suggesting inconsistent locale tagging that will fragment counts. Null rate is 9.67%, and entropy ratio of 0.38 confirms heavy concentration in a few categories.

Holds a film director's name, averaging 2.2 words and 14.8 characters with 62,207 unique values across 143,258 rows. The duplicate rate is 55.3% (76,857 rows), inflated by a 'Unknown Director' sentinel that occurs 3,544 times and should not be treated as a real name. Null rate is 2.93%, and the long tail (David DeCoteau at 129, Sam Newfield at 124) reflects prolific B-movie directors rather than data quality issues.

Holds writer credits, typically one or two personal names averaging 2.7 words and 21 characters, with familiar figures like Jing Wong, Woody Allen, and Ingmar Bergman topping the list. Coverage is weak: 37.1% of rows are null and 25.3% are duplicates across 67,274 unique values, so a single column likely concatenates multiple co-writers per title. Top tokens (michael, david, john) confirm Western personal names dominate, though 'de' hints at multi-name strings or non-English credits mixed in.

Box office gross stored as a short currency string like "$1.1M" — every value is one token, 99.99% allcaps, and lengths cluster between 2 and 7 characters. The column is 89.71% null and only 4,863 distinct values cover the 14,762 populated rows, with a 67.01% duplicate rate concentrated on round million-dollar figures. Note this is a coarse, pre-formatted string (millions only), not a precise revenue number.

This column lists film distributor names, dominated by major studios like Paramount Pictures (994), 20th Century Fox (745), and Universal Pictures (737). It is overwhelmingly sparse with an 83.94% null rate and a 83.94% duplicate rate across 3,694 unique values, suggesting most rows lack distributor data while a small set of studios accounts for the populated entries. Names are short (mean length 19.9 chars, median 2 words) and vocabulary is concentrated around terms like 'pictures', 'films', and 'entertainment'.

Catalogues the audio mix format of each title (Surround, Dolby Digital, Stereo, Mono, etc.), with 551 distinct labels across 143,258 rows. The dominant issue is sparsity: 88.89% of values are null, and even among populated rows 'Surround' covers only 25.6%. Free-form combinations like 'Stereo, Surround' vs 'Surround, Stereo' and overlapping Dolby variants suggest the field is unnormalised multi-label text rather than a clean taxonomy.