Metadata Matching for Piracy Discovery

Every pirated copy arrives wrapped in description. A torrent has a release name and a file list. A cyberlocker file has a name and a size. A streaming index page has a title, a year, a poster, a genre and a cast list. A video upload has a title, tags and a duration. Metadata matching compares that description with what the rights holder knows about its own works, and uses agreement between fields to decide which candidates deserve a closer look.

One title, many spellings

The first job is normalisation. The same film might appear under its original title, an English title or a local one, with or without the leading article, with punctuation stripped, with the year in brackets, and with words separated by dots or underscores. Series episodes are written as season and episode codes, as the word "episode" plus a number, as an absolute episode count from the start of the show, or as an air date.

Normalisation reduces these to comparable forms. Text is lower-cased, punctuation removed, articles handled consistently, alternative titles mapped to one canonical record, and episode references converted to a single scheme. Without that step, matching turns into a long list of hand-written search terms that never quite covers what is out there.

Fields that corroborate each other

Any single field is weak. Titles are shared, years are approximate, and cast names appear on fan pages and reviews. Strength comes from combinations. A candidate whose title is a known variant, whose year is right, whose runtime is close to the reference and whose file size is plausible for a full feature is far more likely to be the work than one that matches on title alone.

Matching systems usually express this as a score built from weighted fields. Rare fields carry more weight: a close runtime says more than a matching genre. Contradictions count too. A candidate with the right title but a runtime of a few minutes is a trailer or a clip, and should be routed differently from a full copy.

Torrent metadata is especially rich. A torrent file lists every file with its name and size, and both a torrent file and a magnet link carry an info hash that identifies that exact set of files. Once one release has been confirmed, the same info hash appearing elsewhere points to the same files without any further analysis.

Remakes, namesakes and trailers

Most false matches come from works that legitimately share descriptive fields. A remake shares its title with the original and differs mainly in year and cast. Two unrelated films can carry the same title. A documentary about a film carries its name. A trailer or a featurette has both the title and the year. Fan edits, recuts and parodies borrow everything.

These cases are resolved by the fields that differ, which is why the reference record needs more than a title. Production year, director, principal cast, runtime, and identifiers from public film and television databases where they exist all help separate the work from its neighbours. The system should also know which related items are authorised, such as official trailers and featurettes, so they are not flagged in the first place.

Building the reference record

Matching quality depends on the reference record. For each protected work it should hold:

• every official title by territory, and the original title in its own script
• common unofficial variants and transliterations
• the year and the runtime of each cut
• principal cast and crew names
• the episode list, for a series
• any public identifiers

Release group names and quality tags seen on earlier pirated copies can be added as they are discovered, because they recur across a title's lifetime. The release group entry explains why those names are such a dependable signal.

Metadata finds the file, content decides

An uploader can rename a file in seconds. Metadata can also be wrong by accident: a mislabelled upload, a fake file named after a popular release to attract downloads, or malware dressed up as a film. Metadata matching should therefore build candidate lists, not file notices by itself.

Confirmation comes from the content: a fingerprint comparison, a visual check, or an analyst opening the stream. Once a candidate is confirmed, its metadata becomes useful again, because the same release name, info hash or file size appearing elsewhere is now a reliable lead to the same copy.