Named entity recognition
Named entity recognition is a language AI technique that finds and labels the names of real-world things in text, such as people, organizations, places, products and dates. In video workflows, named entity recognition runs on transcripts, captions and on-screen text, so a video can be tagged with who and what it mentions. Each entity is linked to the timestamp where it appears.
Why it matters
Names are what people search for: a politician, a company, a product. Tagging videos by the entities they mention makes archives searchable, powers topic pages and enables brand mention tracking without editors logging every name by hand. That makes tagging a full archive realistic: named entity recognition on FastPix costs $0.0035 per minute of video.
How it works
- The video is transcribed, and on-screen text can be added with OCR.
- A language model scans the text for spans that look like names, places, organizations or dates.
- Each span is labeled with an entity type, such as person or organization.
- Entities are stored with timestamps so a search can jump to each mention.
What are examples of named entities?
Named entities are specific, named things in text. Common types are people (a CEO's name), organizations (a company or team), locations (a city or country), products, events, dates and amounts of money. For example, in "the mayor of Chicago spoke on Monday," Chicago is a location and Monday is a date.
How is named entity recognition used in video?
In video, named entity recognition runs on transcripts, captions and on-screen text to tag each video with who and what it mentions, and when. Editors can then search an archive for every clip mentioning a person, and products can build topic pages, track brand mentions or link videos to related articles automatically.
Where it's used
Newsrooms use named entity recognition to tag footage with the people, companies and places mentioned, speeding up archive search. Creator platforms detect brand mentions, and SaaS call tools build searchable records.
Where it usually goes wrong, and how FastPix handles it
Entity extraction is only as good as the transcript beneath it, and running the two in separate tools makes it hard to trace a name back to its moment.
Entities are extracted from each video's transcript by FastPix's in-video AI and returned alongside the transcript itself, summaries and chapters for the same video. Since names come from that transcript, an editor can trace any name back to the video it came from and check it before publishing.
